Rockstar Consortium US LP et al v. Google Inc

Filing 220

MOTION to Amend/Correct Invalidity Contentions by Google Inc. (Attachments: # 1 Affidavit of Lance Yang, # 2 Exhibit 1, # 3 Exhibit 2, # 4 Exhibit 3, # 5 Exhibit 4, # 6 Exhibit 5, # 7 Exhibit 6, # 8 Exhibit 7, # 9 Exhibit 8, # 10 Exhibit 9, # 11 Exhibit 10a, # 12 Exhibit 10b, # 13 Exhibit 11)(Perlson, David) (Additional attachment(s) added on 10/27/2014: # 14 Text of Proposed Order) (ch, ).

EXHIBIT 10b

Reference

Disclosure
1507, and the “Quarks” article 1519 are both science and
technology related. The other two articles 1515 and 1511 are
not. Each article and advertisement contains information that
can be categorized in multiple ways. This categorization
includes at least one topic classifying the information. These
topics are developed and maintained by the information
provider. Using gaze tracking, the information provider can
determine the user’s interest in each displayed article 1505,
1509, 1513 and 1517 and advertisement 1521. Then, by using
the topics categorizing the presented information, the
information provider can dynamically adjust the selection of
subsequent information presented to this user. In the example
above, suppose the user read the scientific based articles 1507
and 1519 but did not spend any time reading the other articles
1511 or 1515 or the advertisement 1521. The information
provider populates the next page of information presented to
the user with articles and advertisements that have similar
topics as the previously read information.
TOGNAZZINI, 16:44-60:
FIG. 16 illustrates a possible second page of information.
Again, the information is provided within views contained in a
window 1601. Now a plurality of articles 1607, 1611, 1615
and 1619 are all scientific or technology based, but with
different levels of difficulty extending from articles of interest
to the lay reader to those that are directed toward the advanced
elemental particle physicist. Further, both a Major Scientific
Headline 1603 and an advertisement 1621 can be selected to
be of interest to the user. This allows the information provider
to narrowly target advertising and articles to each user. Again
the information provider can continue to refine and narrow the
selection of information presented to the user on subsequent
pages depending on the interest shown in a plurality of article
titles 1605, 1609, 1613, 1617, the time spent with reading each
article 1607, 1611, 1615 and 1619; and the time spent looking
at the advertisement 1621 of the current page.
TOGNAZZINI, 16:61-17:12:
FIG. 17 illustrates the process used to select information for
presentation to a user. The process starts at a terminal 1701
after an initial selection of information is displayed to the user.
Using the gaze position developed as described above, a step
1705 monitors the user’s reading pattern. Further a step 1709,
determines the amount of interest shown by the user in the
displayed information. This interest is determined by
measuring the user’s reading speed, determining whether the
user only skimmed the information or read the information in
198

Reference

Disclosure
depth, and by measuring the amount of time spent with each
article and advertisement. Then in a step 1711, the process
retrieves the topics associated with each displayed information
and in a step 1713 correlates these topics with the user’s
interest. Next in a step 1715, the process selects additional
information based on this correlation. This selection of
information is displayed in a step 1717 for the user. Finally,
the process completes through a terminal 1719. In this manner,
the user is presented with a customized set of information that
reflects the user’s interest.
TOGNAZZINI, Fig. 15:

TOGNAZZINI, Fig. 16:

199

Reference

Disclosure

TOGNAZZINI, Fig. 17:

200

Reference

Disclosure

Kamba, Bharat, and
Albers, The Krakatoa
Chronicle – An
Interactive,
Personalized
Newspaper on the Web
(“KAMBA”)

See e.g., KAMBA, p. 1 (“we describe an experimental system which
implements an interactive, personalized newspaper on the WWW.
Some of the parameters for personalization are computed at the server
end, based on user profiles and the composition of the newsfeed.
Personalized layout happens at the client end, based on other
parameters under user control.”); id., p. 2 (“A user’s profile is
modified by the explicit feedback provided by the user on the
relevance of various articles, and when this is unavailable, from
implicit feedback, derived from observations made by the embedded
Java agent. The agent observes the manner in which the user interacts
with the articles in the document, and based on the time spent, the
interaction techniques used (e.g. scrolling, peeking at, maximizing,
resizing), it tries to estimate the user’s interest and modifies the user’s
profile suitably.”); id., p. 4 (“The weight of each keyword represents
the system’s reckoning of the user’s interest in the keyword. It is
computed when feedback is given. Feedback provides a score for the
201

Reference

Edwards, Bayer, Green
& Payne, Experience
with Learning Agents
which Manage
Internet-Based
Information, AAAI
Technical Report SS96-05, 1996
(“EDWARDS”)

Lieberman, Letizia: An
Agent That Assists Web
Browsing
(“LIEBERMAN”)

Disclosure
whole article which is then used to compute scores for individual
keywords in its document-vector. Then it is integrated into the user’s
profile.”); id., p. 7 (“Layout is a function of several parameters: the
score that each article receives based on the user’s profile (the user
score), the average score received by each article over the community
of users (the community score), and also the size and composition of
each article . . . The order of articles . . . for a user is decided by each
article’s score, and the score is a function of the user’s score and the
community score.”); id., p. 8 (“all these interactions give feedback
about the relevance of the article to various degrees . . . When the user
scrolls, peeks at, maximizes, resizes, or saves an article to a scrapbook,
the Krakatoa Chronicle increments the user’s interest in the article by a
corresponding amount, and subsequently changes the personal
profile.”)
See e.g., EDWARDS, p. 31 (“An alternative solution is to build a profile
which reflects the user’s preferences when using an application, such
as a World-Wide Web browser.”); id., p. 33 (“The architecture can be
divided into two broad areas: the Profile Generation Phase and the
Classification/Prediction Phase. The Profile Generation phase is
responsible for inducing the user profile . . . Actions performed by the
user on a document (news article, Web page, etc.) are recorded
together with the text of the document. Features are extracted from
these observations, and used to create a training instance. The training
instances are then used to induce the user profile. . . . The
Classification/Prediction phase is responsible for determining the
actions to be performed on new documents. . . . Features are extracted
from each document, and the user profile employed to generate a
classification (with an associated confidence rating). The confidence
rating is used by the Prediction Stage to determine whether a
prediction should be made.”); id., p. 35 (“LAW (Bayer 1995) is a
system that helps a user find new and interesting information on the
World-Wide Web. It provides assistance in two ways: by interactively
suggesting links to the user as they browse the Web; and through the
use of a separate Web robot that autonomously searches for pages that
might be of interest.”).
See e.g., LIEBERMAN, p. 2 (“This paper introduces an agent, Letizia,
which operates in tandem with a conventional Web browser such as
Mosaic or Netscape. The agent tracks the user’s browsing behavior –
following links, initiating searches, requests for help – and tries to
anticipate what items may be of interest to the user. It uses a simple
set of heuristics to model what the user’s browsing behavior might
be.”); id. (“Letizia uses the past behavior of the user to anticipate a
rough approximation of the user’s interests.”); id., p. 3 (“The goal of
the Letizia agent is to automatically perform some of the exploration
that the user would have done while the user is browsing these or other
202

Reference

Disclosure
documents, and to evaluate the results from what it can determine to
be the user’s perspective.”); id. (“One of the strongest behaviors is for
the user to save a reference to a document, explicitly indicating
interest. . . . Following a link is, however, a good indicator of interest
in the document containing the link. . . . Repeatedly returning to a
document also connotes interest . . . a link that has been ‘passed over’
can be assumed to be less interesting.”); id., p. 4 (“it is Letizia’s job to
recommend which of the several possibilities available is most likely
to satisfy the user.”); id., p. 6 (“the agent serves the role of
remembering and looking out for interests that were expressed with
past actions.”).
Lam, Mukhopadhyay,
See e.g., LAM, p. 317 (“Information filtering is concerned with the
Mostafa, and Palakal,
problem of delivering useful information to a user while preventing an
Detection of Shifts in
overload of irrelevant information. Information selected for
User Interests for
presentation is commonly based on descriptions of user preferences
Personalized
called profiles. Typically, the user profile is not known in advance,
Information Filtering,
and can also change with time. The user may choose to provide a
SIGIR’96, ACM 1996
limited amount of feedback information concerning the relevance of
(“LAM”)
specific items. The objective is to estimate the user profile from the
feedback data so that the filtering system can effectively choose and
present information as relevant to the user as possible.”); id., (“in the
case of text-based document filtering, the overall problem of
information filtering may be broadly posed as learning a map from a
space of documents to the space of real-valued user relevance factors. .
. . a finite set of documents can always be rank-ordered and presented
in a prioritized fashion to the user.”); id., p. 318 (“a user profile
learning module that learns user interests over the document
categories, based on on-line user relevance feedback and a
reinforcement machine learning algorithm.”); id., p. 320 (“The user
profile learning module consists of a learning agent that interacts
directly with the user and sorts the incoming documents according to
its belief of the user preferences for the various categories of
documents. To accomplish this task, the learning agent maintains and
updates a simplified model of the user.”).
See e.g., O’RIORDAN, p. 205 (“We present here an overview of a
O’Riordan and
Sorensen, An Intelligent research project aimed at reducing information overload for individual
Agent for Highcomputer users.”); id. (“It is generally acknowledged that the volume
Precision Text
of information which is accessible over various networks has exceeded
Filtering, CIFM ’95,
the capability of users to sift through it in order to access that which is
ACM 1995
relevant to them.”); id. (“an information filter was built which can be
(“O’RIORDAN”)
personalized by individual users and which models the user’s interests
so as to route through to him/her those articles which are deemed as
relevant. The user may evaluate the significance of received
information, thus providing relevance feedback which is used in finetuning the filter (or user profile) so as to improve its precision and to
203

Reference

Bloedorn, Mani,
MacMillan, Machine
Learning of User
Profiles:
Representational
Issues, Proceedings of
AAAI-96, Portland,
OR, Aug. 4-8, 1996
(“BLOEDORN”)

Pazzani, Muramatsu, &
Billsus, Syskill &
Webert: Identifying
interesting web sites,
AAAI 1996
(“PAZZANI”)

Maes, Agents that

Disclosure
better model a user’s changing interests. In this sense, the profile
learns of a user’s preferences through assimilation of an initial set of
interesting documents and continues this learning process via
relevance feedback throughout its lifetime.”); id., p. 206 (“The basic
assumption is that a software agent acts on behalf of the user –
embodying his/her beliefs, intentions and goals – behaving as an
intermediary between the user and the system with which he/she is
interacting.”); id., p. 208 (“The comparison of a user profile with a
document representation involves the localized matching of structural
similarity between the profile network and incoming article networks,
using profile weights to influence this comparison.”); id., p. 209
(“Those articles considered relevant to the user’s needs are forwarded
by the agent, while the others are screened out. Forwarded articles are
also ranked according to estimated relevance.”); id. (“Via the user
interface, the user may provide relevance feedback on those articles
routed to him/her.”).
See e.g., BLOEDORN, p. 1 (“The goal of the research described here is
to build a system for gathering comprehensible user profiles that
accurately capture user interest with minimum user interaction.”); id.,
p. 2 (“Our experiments were conducted in the context of a contentbased profiling and summarization system for on-line newspapers on
the World Wide Web, the IDD News Browser. In this system, the user
can set up and edit profiles, which are periodically run against various
collections built from live Internet newspaper and USENET feeds, to
generate matches in the form of personalized newspapers. These
personalized newspapers provide multiple views of the information
space in terms of summary-level features. When reading their
personalized newspapers, users provide positive or negative feedback
to the system, which are then used by a learner to induce new profiles.
These system-generated profiles can be used to make
recommendations to the user about new articles and collections.”)
See e.g., PAZZANI, p. 1 (“In this paper, we discuss Syskill & Webert, a
software agent that learns a profile of a user’s interest, and uses this
profile to identify interesting web pages in two ways. First, by having
a user rate some of the links from a manually collected ‘index page’
Syskill & Webert can suggest which other links might interest the
user. . . . Second, Syskill & Webert can construct a LYCOS query and
retrieve pages that might match a user’s interest, and then annotate this
result of the LYCOS search.”); id., p. 3 (“The user profile is learned
by analyzing all of the previous classifications of pages by the user on
this topic. If a profile exists, a new profile is created by reanalyzing
all previous pages together with any newly classified pages. Once the
user profile has been learned, it can be used to determine whether the
user would be interested in another page.”).
See e.g., MAES, p. 32 (“The machine learning approach is inspired by
204

Reference
Reduce Work and
Information Overload,
Communications of the
ACM, July 1994
(“MAES”)

Sheth and Maes,
Evolving Agents for
Personalized
Information Filtering,
1993 IEEE (“SHETH”)

Disclosure
the metaphor of a personal assistant. Initially, a personal assistant is
not very familiar with the habits and preferences of his or her
employer and may not even be very helpful. The assistant needs some
time to become familiar with the particular work methods of the
employer and organization at hand. However, with every experience
the assistant learns, either by watching how the employer performs
tasks, by receiving instructions from the employer, or by learning from
other more experienced assistants within the organization. Gradually,
more tasks that were initially performed directly by the employer can
be taken care of by the assistant. The goal of our research is to
demonstrate that a learning interface agent can, in a similar way,
become gradually more helpful and competent.”); id., p. 33 (“the
interface learns by continuously ‘looking over the shoulder’ of the user
as the user is performing actions.”); id. (“A second source for learning
is direct and indirect user feedback. Indirect feedback happens when
the user neglects the suggestion of the agent and takes a different
action instead. This can be as subtle as the user . . . not reading some
articles suggested by the agent . . .”); id., p. 34 (“the agent can learn
from examples given explicitly by the user.”); id., p. 38 (“A user can
create one or many ‘news agents’ and train them by means of
examples of articles that should or should not be selected. . . . The user
can also program the agent explicitly and fill out a set of templates of
articles that should be selected (e.g., select all articles by Michael
Schrage in the Los Angeles Times). Once an agent has been
bootstrapped, it will start recommending articles to the user. The user
can give it positive or negative feedback for articles or portions of
articles recommended.”); id., p. 39 (“The agents are able to
recommend articles to the user that concern topics (or authors or
sources) in which the user has shown a continued interest.”).
See e.g., SHETH, p. 340 (“Filtering system can be viewed as a search
process. It involves searching over the large and complex space of
possible user profiles, for an ‘optimal’ user profile (or set of profiles)
that match the user’s different interests. This ‘optimal’ user profile
has to vary as the user’s interests change over time.”); id. (“The
system consists of a number of news categories which a user has
defined. Each of these news categories consists of a population of
filtering agents. These are ‘organisms’ that retrieve articles which
match an internal representation of the type of article they are
interested in. The internal representation consists of whatever the
organism inherited generically from its parents (the genotype)
augmented with information it learns during its lifetime. Agents are
assigned a fitness value based on the user feedback regarding their
performance. The user conveys whether an article that was retrieved
by one or several agents was appreciated or not. The agents learn
from this feedback by changing their internal representation to reflect
205

Reference

Balabanovic, An
Adaptive Web Page
Recommendation
Service, 1997 ACM
(“BALABANOVIC”)

Disclosure
this training example. For each positive/negative feedback received,
an agent gets positive/negative fitness points.”); id., p. 347 (“When an
agent receives positive feedback, it extracts information from the
corresponding article and incorporates it into its internal
representation. Presently, the agent extracts most of the information
provided in the header of the news article (Figure 1), in particular the
author, keywords, location, category and priority fields. If, say, a
keyword is already present in the internal representation, it’s weight is
increased, so that the agent is more likely to retrieve similar articles in
the future. Conversely, in the case of negative feedback, the
information is stored with negative weight, so as to make it less likely
that similar articles will be retrieved in the future. The user can also
manually indicate preference for particular keywords occurring in an
article.”); id., p. 349 (“The user can give positive or negative feedback
by clicking on the ‘thumbs-up’ or the ‘thumbs down’ icon
respectively.”).
See e.g., BALABANOVIC, p. 378 (“In this paper we introduce the ‘Fab’
adaptive Web page recommendation service. . . . Running since March
1996, it has been populated with a collection of agents for the
collection and selection of Web pages, whose interaction fosters
emergent collaborative properties.”); id. (“The operation of the system
is as follows: users can request recommendations at any time, and will
be shown the ten highest-ranking Web pages according to their
profile.”); id., p. 380 (“All agents maintain a profile: each user has a
selection agent, which maintains their user profile; each collection
agent maintains a search profile which is used to guide it in its
collection of web pages.”); id. (“At regular intervals collection agents
submit the pages they have gathered which bets match their search
profiles to the central repository, replacing the pages they previously
submitted. Thus at any time the repository contains each collection
agent’s best pages (in their own opinions). When a user requests their
Fab recommendations their selection agent (of which there is one per
user) picks, from the entire repository, those pages which best match
the user’s personal profile. The user then rates these pages. These
ratings are used as feedback for the agents to learn from, and the
resulting rankings are used for evaluation purposes (discussed in
section 4). The selection agent uses the feedback to update the user’s
personal profile (using the function u). It also forwards the feedback,
via the central repository to the originating agent A, which will update
its search profile in the same way.”); id. (“A brand new user to the
system is shown a selection of pages which are randomly chosen from
the repository. However the repository contains pages which various
agents believe will best match the current user population. Thus the
new user is already starting from a much higher level than would be
expected from an empty profile, especially if the system is deployed in
206

Reference

Disclosure
an organization or special interest group where there will be
significant overlap between users’ interests.”); id., p. 381 (“Rather
than actually searching the Web, these agents attempt to construct
queries for existing Web indexes in an attempt to avoid duplicating
work. The indexes used are Alta Vista, Inktomi and Excite.”); id., p.
382 (“The highest-scoring pages are shown to the user, with the
proviso that no two are identical or from the same site, and that the
user has not seen an identical page in the last month.”); id. (“On a dayto-day basis the system supplies the user with a number of documents
it thinks the user will rate highly. It uses the resulting scores in order
to perform relevance feedback and improve the user profile.”)
Fox, Hix, Nowell,
See e.g., FOX 1993, p. 485 (“The Query Window has two categories of
Brueni, Wake, and
use: . . . Access to previously completed (old) queries and the results
Heath, Users, User
of the related searches are provided. Old queries may simply be
Interfaces, and Objects: viewed or they may be revised and used for another search. Results of
Envision, a Digital
searches from old queries may also be redisplayed via a query history
Library, Journal of the feature.”); id. (“As queries are stored or related searches are
American Society for
performed, the user establishes a history that is accessible through the
Information Science,
Query History field across the top of the window . . . The Query
44(8):480-491, 1993
History provides access to the results of previous searches, means to
(“FOX 1993”)
redisplay the full content of previous queries for possible revision, and
a mechanism for combining the results for completed searches.”)
Little, Commerce on
See e.g., LITTLE, p. 75 (“On-line services can incorporate customer
the Internet, 1994 IEEE preferences and use history, such as past purchases or chapters read, to
(“LITTLE”)
provide a personal environment to the customer, saving access time.”);
id., p. 76 (“For example, a customer might, through an interactive
form, indicate current age, number of children, expendable income,
and home value to identify investment options for a mutual fund
buying service. Using that information, the service might steer the
customer to performance indices that help in choosing investments.”);
id. (“Once interaction is supported, data on individuals can be
maintained both by direct customer involvement (for example,
updating the name and mailing address) and by monitoring the
documents accessed. A personal profile can capture basic
demographics as well as individual information and environmental
preferences. This information can be used for a number of interesting
purposes, including 1. to configure the interface presentation 2. to fuel
Web ‘agents’ who actively search the net or site based on the profile,
and 3. to tailor and select site-specific information to present to the
customer (for example, showing children’s ads to children and adultoriented ads to adults).”).
Adam and Yesha,
See e.g., ADAM, p. 822 (“From a consumer’s perspective, EC/DL
Strategic Directions in systems require decision agents that can learn an individual
Electronic Commerce
consumer’s preferences, seek out appropriate providers and negotiate
and Digital Libraries:
requests for further information (e.g., to bring to the user’s attention)
207

Reference
Towards a Digital
Agora, ACM
Computing Surveys,
Vol. 28, No. 4, Dec.
1996 (“ADAM”)
U.S. Patent No.
5,933,811 to Angles et
al. (“‘811 PATENT”)

Disclosure
or initiate purchases.”)

‘811 PATENT at Abstract, “The present invention is a system and
method for delivering customized electronic advertisements in an
interactive communication system. The customized advertisements are
selected based on consumer profiles and are then integrated with
offerings maintained by different content providers. The preferred
interactive communication system interconnects multiple consumer
computers, multiple content provider computers and multiple Internet
provider computers with an advertisement provider computer.
Whenever a consumer directs one of the consumer computers to
access an offering existing in one of the content provider computers,
an advertising request is sent to the advertisement provider computer.
Upon receiving the advertising request, the advertising provider
computer generates a custom advertisement based on the consumer's
profile. The custom advertisement is then combined with the offering
from the content provider computer and displayed to the consumer.
The advertisement provider computer also credits a consumer account,
a content provider account and an internet provider account each time
a consumer views a custom advertisement. Furthermore, the
advertisement provider computer tracks consumer responses to the
customized advertisements.”
‘811 PATENT, e.g., Col. 2, “As the popularity of the Internet and the
World Wide Web has increased over the years, more companies are
trying to find ways of promoting their product in a cost-effective
manner. Thus, there has been a tremendous proliferation of corporate
advertising across the Internet. For example, some companies such as
Yahoo Corporation offer free services, such as the ability to search for
particular sites on the Internet, but post advertising messages to
consumers to help offset the cost of their service. Unfortunately, there
is so far no effective way of targeting particular advertisements to
those consumers most likely to use the product or service being
offered. Therefore, a tremendous amount of advertising is wasted on
promoting goods or services to an improper audience. As the number
of people accessing the Internet increases, it will become more
important to specifically target advertising to those individuals most
likely to purchase the goods or services being offered. It will also be
important for advertisers to know how effective a particular ad has
become by tracking the responses of individual consumers.
Unfortunately, there is currently no convenient mechanism for
predetermining which users might be interested in a particular
208

Reference

BROADVISION

Disclosure
category of advertised goods or services. There is also no current
method for tracking consumer responses to particular advertisements.”
‘811 PATENT, e.g., Summary of the Invention; Claims 1, 4, 6, 12;
Figures 1-11
Press Release (1.22.96)13, e.g., “One, the first application system for
dynamic personalized marketing and selling on the Internet's World
Wide Web. Over two years in development, the BroadVision One-ToOne software product transforms static "brochureware" Web sites into
interactive, one-to-one marketing communities. These online
communities, built around consumer brands, virtual malls, or valueadded services, will enable businesses to build long-term relationships
with their customers through personalized content, services and
promotions. Using the product's innovative Dynamic Command
Center feature (for which the company has a pending patent),
marketing, advertising and Web content managers, can: Personalize
editorial content, advertising and incentive programs based on
individual consumer demographics, psychographies and usage
patterns; Observe consumer interactions in real time to identify and
seize opportunities based on understanding and responding to
consumers' online activity; Foster virtual communities by easily
integrating electronic mail bulletin boards and online forums to OneTo-One applications; Establish collaborative online dialogues with
customers to improve long-term satisfaction and retention.
Press Release (1.22.96), e.g., “According to Don Peppers, a wellknown marketing consultant and co-author of the best-selling book
"The One-To-One Future," the most important challenge facing
marketers today is to build life-long customer relationships. "But to
achieve this goal and realize the enormous potential of the Web,
marketers need more than cool graphics and secure transactions. To
keep consumers coming back, Web sites must 'learn' from interactions
and remember from visit-to-visit the unique preferences and interests
of each individual," Peppers said. "Savvy Web marketers will use
sophisticated software like BroadVision One-To-One to progressively
enhance the quality of information exchange with their customers,
resulting in strong one-to-one relationships that deliver increasingly
greater benefits to both producer and consumer."”

13

PRESS RELEASE (1.22.96) shall refer to Personalized Marketing and Selling on the Internet
Unleashed by BroadVision One-To-One Application System Helps Marketers Build Long-Term
Relationships Through Personalized Content, Services and Promotions,” dated Jan. 22, 1996.

209

Reference

Disclosure
Press Release (3.21.96)14, e.g., “Broadvision Inc is to offer
personalised Web sites based on individual user profiles with its Oneto-One on-line marketing database. One-to-One, which has been two
years in development, offers a three-tier environment enabling
businesses to tailor their World Wide Web sites to individual
customers by tracking their preferences as they move around the site.
In this way, companies can build up very detailed 'psychographic'
profiles of their customers, enabling them to target specific advertising
and promotions to individuals.”
Press Release (3.21.96), e.g., “Chen admits that interactive,
personalised marketing is not new, but says that the Internet is
enabling it to be practised on a very large scale. Broadvision is looking
at the complete life-cycle, from attracting customers to the site,
encouraging them to buy over the Internet, offering incentives to give
their details to the company and finally supporting the actual payment
transaction.”
Press Release (5.15.95)15, e.g., “BroadVision Inc. today proposed the
definition of a new category of software critical to manage the buying
and selling of products and services
via interactive networks. The new category, Interactive Commerce
Management System (ICMS), is a comprehensive solution for
electronic commerce that allows interactive service providers to
conduct interactive marketing, ordering and billing online.
BroadVision, based in Los Altos, Calif., is currently designing and
developing the first ICMS product for delivery by year-end 1995.”
Press Release (5.15.95), e.g., “An ICMS employs object technology to
enable traditional database marketing and new interactive marketing
practices, including tracking consumer usage and interests. Service
providers can evaluate the effectiveness of a particular promotion to
reward repeat customers or determine peak shopping times. Since an
ICMS uses objects to represent business rules and processes, it can
reflect the unique business models of the service providers and still
respect the specific 'look and feel' of the storefronts, allowing for

14

PRESS RELEASE (3.21.96) shall refer to “Broadvision Uses Its On-Line Marketing Database
To Deliver Personalised World Wide Web Sites,” dated Mar. 21, 1996.

15

PRESS RELEASE (5.15.95) shall refer to “BroadVision Developing First Interactive Commerce
Management System To Support Online Sales & Marketing Process; New Software Category
Necessary to Interactive Network Architecture,” dated May 15, 1995.

210

Reference

Disclosure
change over time.”
Press Release (5.15.95), e.g., “Examples of electronic commerce
applications that can be offered to consumers over an interactive
network include full-service electronic malls, independent electronic
retailing, personalized advertising, travel services, movies-on-demand,
time-shifted TV, pay-per-view, automated ticket sales, educational
programs and online games.”

C/NET

Press Release (12.18.95)16, e.g., “C/net unveiled a new system that
allows advertisers to target narrow audiences by delivering different
ads to different site visitors in real time. The technology, called
DREAM (Delivery of Real-Time Electronic Advertising Messages),
went into operation on c/net's two web sites Dec. 15. DREAM allows
c/net to categorize visitors to its site based on demographic
information (taken from site registrations) and hardware and software
data gathered on the fly. “We know certain things about the people
coming into our site,~ said Scott Waltz, c/net's vice president of
marketing. ~we know what kind of platform they're coming with, their
connection rate, their browser type, and so on. We use that information
to affect how our database serves content.” DREAM allows an
advertiser to display different banners to different users. Waltz
explained, "If I know someone's coming in with a Mac instead of an
Intel platform, I can tailor the software or peripherals that I offer to
that person so I have a much higher chance of speaking to that person
and offering them something that they want." In another example, a
bank could present a standard credit card offer to all customers except
those from .edu domains. Visitors from .edu -- many of which are
college students -- would see a special first card offer.”

APTEX

www.aptex.com17, e.g., “Aptex provides text analysis software to
enhance mission-critical, real-time business processes and decisions.
Using proprietary Content Mining™ technology, Aptex develops and
markets intelligent solutions for online publishing, market intelligence,
customer response, and educational publishing. Aptex products
include Convectis™, an intelligent document categorization and
routing server, and VITAL ResourceMiner™, an interactive tool for
correlating educational content to state and local instructional

16

PRESS RELEASE (12.18.95) shall refer to “Online Marketing: C/Net Introduces Customized
Web Advertising,” dated Dec. 18, 1995.

17

WWW.APTEX.COM

refers to webpages accessed through www.aptex.com homepage, available
through www.archive.org (last accessed on May 19, 2014).

211

Reference

Disclosure
standards. SelectCast™, an intelligent advertising and audience
management server for World Wide Web content providers, is
scheduled for delivery in the second half of 1996.”
www.aptex.com, e.g., “SelectCast is a Web advertising placement
server with unique predictive modeling capabilities for increased
advertising effectiveness. When integrated with a Web site, the
SelectCast advertising server will present visitors with intelligentlyplaced, individually-tuned advertising and promotions. Using
proprietary Aptex neural network techniques and Content Vector™
data model, SelectCast will develop self-adjusting, predictive
models of user behavior. By correlating these user profiles with
advertising performance, demographic databases, and content provider
feedback, SelectCast will continually improve advertising placement
effectiveness.”
Press Release (5.6.96)18, e.g., “Infoseek Corporation, a leading Web
search service, and Aptex, a newly formed division of HNC Software
Inc., today announced a long-term development and marketing
partnership. Under the terms of the agreement, HNC's Aptex Division
and Infoseek will jointly design and market SelectCast™, an
intelligent advertising and audience management server for the World
Wide Web, based on HNC's text analysis technology. Infoseek will
also use the Aptex division's Convectis™ product, a neural networkbased text analysis server, to automatically update the directory
portion of Infoseek Guide, Infoseek Corporation's flagship Internet
service. . . When integrated with Guide, SelectCast is designed to
present visitors with intelligently disseminated, individually targeted
advertising and promotion. Using proprietary neural network
techniques and the HNC patented Context Vector™ data model,
SelectCast is intended to deliver self-adjusting, predictive models of
user behavior. . . To expand and enrich the user's Web experience,
Infoseek will also employ the HNC/Aptex Convectis server as an
"intelligent librarian" aide to its experts who categorize and summarize
Web pages into thousands of categories within Infoseek's Guide
directory. Convectis is expected to allow Infoseek's Guide directory to
scale continually with the growth of the World Wide Web.”

18

PRESS RELEASE (5.19.14) refers to “HNC Software and Infoseek Announce Web
Partnership,” dated May 6, 1996.

212

Reference

Disclosure
Press Release (12.3.96)19, e.g., “Aptex Software Inc. Tuesday
announced the availability of SelectCast for Ad Servers, an intelligent
software solution that revolutionizes Internet advertising by
maximizing ad clickthrough and selectively targeting specific
audiences.
Press Release (12.3.96), e.g., “"Online advertisers are asking for
higher response rates and more audience selectivity," said Michael
Thiemann, President and CEO of Aptex. "SelectCast for Ad Servers
delivers on both counts, with what we believe are the highest sustained
clickthrough rates and most selective ad targeting available on the
Internet." Aptex developed SelectCast for Ad Servers in partnership
with Infoseek Corp. (NASDAQ: SEEK) to deliver industry-leading
capabilities for a new Infoseek advertising service currently in
development. Infoseek expects to use SelectCast technology
throughout its service and the Infoseek Network. Early versions of
SelectCast have been in use at Infoseek since the summer of 1996.
"SelectCast capabilities represent the state of the art and a major
improvement in ad serving technology," said Robin Johnson, Infoseek
CEO. "We selected Aptex as a strategic partner because we believe
their software is far superior to other technologies we evaluated for
personalization and ad serving."”
Press Release (12.3.96), e.g., “SelectCast for Ad Servers improves
clickthrough rates by continuously evaluating user profiles as users
click ads, and then delivering the same ads to similar users. SelectCast
for Ad Servers delivers clickthrough increases of up to 50 percent
compared to word- and topic-matching selection techniques -- until
now the bestperforming technology available -- and up to 25 percent
when measured against aggregated matching and general rotation
results. SelectCast for Ad Servers targets audiences by developing
profiles for all site visitors, analyzing and grouping profiles to identify
users with similar interests, and then delivering designated ads
consistently to users in selected groups. This "affinity modeling"
process also identifies new audiences automatically as they emerge.
SelectCast for Ad Servers provides comprehensive, site-wide user
profiling while maintaining persistent, multi-visit profiles for every
site visitor, and updating these profiles immediately with every user
action. SelectCast for Ad Servers acts as an "intelligent observer,"

19

PRESS RELEASE (12.3.96) refers to “Aptex announces SelectCast "turbocharger" for
advertising servers,” dated Dec. 3, 1996.

213

Reference

Disclosure
mining the context and content of all actions -- including clicks,
queries, page views and ad impressions -- so that no explicit user
feedback or "taste" judgments are ever required. By irreversibly
encoding all user profiles, SelectCast for Ad Servers ensures user
privacy. Personal information is never requested or stored, and profiles
cannot be reverse engineered to determine specific user actions.
SelectCast for Ad Servers is based on Aptex's patented Content
Mining technology, which employs neural networks and a Context
Vector data model to optimize relationships between users and
content. Future SelectCast products are expected to enhance the
performance of other types of commercial servers - including those for
electronic commerce, one-to-one marketing, online publishing, and
community creation -- by personalizing the selection of product and
service information, news and entertainment, forums and chat
sessions, and other forms of content.”

HYPER-TARGETED
MARKETING

Press Release (12.4.96)20, e.g., “Hyper-Targeted Marketing precisely
targets marketing efforts based on user profiles and choices made
while browsing a Web site. Hyper-Targeted Marketing is based on
Alpha Base Interactive's Metropolis Database and Web Hypertext
Applications Processor (WHAP), a complete system that builds
sophisticated web services that are automatically customized to the
interests of subscribers and customers. Together with the company's
EZ-ID browser plug-in, which automatically provides user
identification, the system keeps track of customer preferences, service
history, and interests they have shown on previous visits to a WHAPsupported web site. "The ability to precisely target marketing efforts is
one of the most compelling advantages to marketing on the Web," said
Steve Fecske, CEO of ABI. "In a world of information overload,
customers will respond best to companies that can match individuals
to products and content of specific interest to them."”

CYBERGOLD

“The CyberGold Service,” e.g., “Upon logging in to the Net, Karen is
presented with a short list of ad titles. Each of them involves a product
or service in which she has actively expressed an interest either
through her previous use of CyberGold facilities or through the user
profile she filled out when she joined CyberGold Today's list contains
ads for medium price hotels in New Orleans (where Karen's family is
plaruting a vacation), a makeit-yourself telescope kit (a possibility for
her husband's upcoming birthday), recently released movies (she's a

20

PRESS RELEASE (12.4.96) shall refer to “Alpha Base Interactive Provides Hyper-Targeted
Marketing Service,” dated Dec. 4, 1996.

214

Reference

Disclosure
fan), some new nonfat organic dessert items (she's on a diet), and
family minivans (with the new baby, the family has outgrown their
present car). Not only are the subjects of the ads keyed to Karen's
interests, but certain aspects of their style and depth are as well This
permits the design of ads that are virtually custom-fitted to Karen's
preferences, thus ensuring her attention The ad messages will be
welcomed and attentively viewed. For the minivan ads, Karen has
requested straightforward technical specs of models and
configurations (she does not need to be sold on the idea of this kind of
vehicle; she already knows she wants one). For the movie ads, Karen
might request a film clip, while another subscriber might ask for a plot
summary. Some consumers might prefer the entertainment value of
ads like those generally found in the mass media, while a subscriber
viewing an ad for food or drink might ask for a list of ingredients or
nutrients.”
“The CyberGold Service,” e.g., “Advertisers will find their potential
customers through patent-pending "demographic routing" technology,
which will steer ads directly to interested and willing buyers, as
defined by the personal profiles in the CyberGold database. A
welcome side-effect of this type of routing is that advertising will
become 'orthogonal,' that is, unlinked to the editorial content of
entertainment and information on the Net. When advertisers aim their
messages at individual consumers rather than at demographic
segments of the population ('blue-collar urban women under 30') they
no longer need to worry about whether the editorial content of a
particular magazine or television show is likely to attract potential
buyers. Advertisers using orthogonal sponsorship typically would not
even know what content they are sponsoring. Instead, they would
simply explicitly delineate their target audience, provide ads, and offer
some form of compensation directly to those viewers willing to view
the ads. This unlinking between advertising and content is likely to be
beneficial to consumers, advertisers, and society.”
“The CyberGold Service,” e.g., “The CyberGold Marketing System is
a more effective way of advertising for four reasons:
* Reach: advertisers can entice more customers to interact with their
advertising by
rewarding customers directly for their attention.
* Targetability: CyberGold makes more efficient use of advertising
dollars by targeting
customer by demographic, psychographic or behavioral
characteristics.
*Accountability: advertisers only pay customers who interact with
their ads. CyberGold
215

Reference

Disclosure
provides information on the characteristics of the customers the ads
have
reached.
* Integration: on-line promotional mechanisms including coupons and
rebates are
encapsulated into a single, easy-to-use system. CyberGold handles the
complexity of
electronic commerce for advertisers.”

FREELOADER

Press Release (9.30.96)21, e.g., “Beginning today, FreeLoader, a unit
of Individual Inc. (NASDAQ:INDV), will make version 2.0 available
for free downloading from the FreeLoader home page
(http://www.freeloader.com). Among the unique features of the new
version are:
- User-defined, custom channels in addition to the "subscriptions" of
version 1.0. These channels allow users to personalize exactly how
they would like to retrieve selected Web content
- "Premium" branded channels of content from popular Web sources
including: MSNBC (www.msnbc.com), NewsPage
(www.newspage.com), Pathfinder (www.pathfinder.com), Slate
(www.slate.com), Sony Music Entertainment (www.sony.com/Music),
Sportsline USA (www.sportsline.com), and ZD Net (www.zdnet.com).
- Enhanced, easier-to-use interface
-Support for Microsoft's Internet Explorer, as well as Netscape
Navigator
- Ability to track user clicks and preferences to offer valuable
advertising and editorial content personalized for each user”
Press Release (9.30.96), e.g., “FreeLoader provides an advertisersupported, free service which automatically retrieves and categorizes
content from pre-selected Web sites at user-defined times. Unlike
other offline services, FreeLoader employs intelligent agenting
technology to passively create a user profile based on clicks and
selections, providing one of the only platforms for advertisers to
customize a marketing message at a specific and well-defined
audience. The server side database keeps track of the statistics
received from the user such as
age group, sex, zip code, country, first name, e-mail address,
occupation and salary group.”

21

PRESS RELEASE (9.30.96) shall refer to “FreeLoader releases Personalized Web Content
Delivered Redesigned Interface And Screen Saver,” dated Sep. 30, 1996.

216

Reference

Disclosure

HYPER SYSTEM

“Hyper System: Patent Pending,” e.g., “What is Hyper System? Hyper
System employs message display application software called "Hot
Cafe'" to transmit advertising and messages to precisely targeted
audience of Internet. Hyper System can run simultaneously. with any
Internet communications application, including WWW browser. It
allows advertisements, information, and messages to be displayed
continuously during a user's dial-up session. This will be forwarded to
the providers via a leased line.”
“Hyper System: Patent Pending,” e.g., “What is Hot Cafe? When
using Hyper System, in addition to the web browser window on a
user's screen, a section of the screen is devoted to another window
"Hot Cafe" where advertising and information messages from
corporations or individuals are continuously displayed. This
information is updated every minute, irrespective of the Internet
communications application. A feature of this "Hot Cafe" gives ·
advertisers the option of incorporating buttons with link capabilities to
their messages, which enable users to easily dick onto the web page of
the advertiser or infomation sender with their web browser. The
application software, "Hot Café”, is distributed to users free of
charge.”
“Hyper System: Patent Pending,” e.g., “Establishing User Profile.
When users register on-line, in addition to providing their names and
addresses, users are asked to complete a detailed questionnaire about
hobbies, interests and so on. Although no personal data is ever
released, the questionnaire responses are plotted to create a statistical
profile of Hot Cafe users. HYPER NET establishes a Database Center in
Japan to match user profiles with the targeted data required by
advertisers. This center will be connected to providers by a leased line
to deliver advertising and information to users. Bendits of Using
Hyper System. Hyper System has benefits for everyone: Providers
have a new source of cash flow, advertisers and infonnation providers
have a new direct marketing tool, and users can save their connection
fees.”
“Hyper System: Patent Pending,” e.g., “Benefit for Advertisers.
Advertisers benefit in several ways using Hyper System. Information
and advertising messages can be targeted precisely at a specific
audience, whose composition is determined by the responses to the
questionnaire. Since the advertising messages are interactive, an
efficient response can be obtained and the results of advertising can be
measured accurately and quickly.”

217

Reference
I/PRO

Disclosure
“About I/Pro,” e.g., “With I/COUNT; site owners can monitor aspects
of site usage such as number of visits, most frequently accessed files,
and geographic and organization
origin of visitors. I/COUNT has been commercially available since
May 15.”
“About I/Pro,” e.g., “I/ CODE: How It Works On Your Site
The Exchange of Value The I/CODE system is based upon a value
exchange between the user and the site. By providing you with their
demographic profile, and potentially their
identity, I/CODE members are sharing very valuable information. For
sites, this demographic data translates into real dollars for I/CODE
enabled sites who can interpolate content, exposure, and advertising
potential from it. Sites should therefore be willing to compensate
I/CODE members for their time and information with give-aways,
sweepstakes, discounts and
other benefits. Free Demographic Data. I/CODE provides sites with
raw demographic data about all I/CODE members who sign-on at their
site absolutely free. If you would like someone to provide data
analysis, I/CODE offers analysis and reporting services on this data
(see I/CODE Reports for more information, but the raw data is
supplied to all participating sites at no cost.”

F.J. Burkowski,
“Delivery of Electronic
News: A Broadband
Application”
(“BURKOWSKI”)

“About I/Pro” (5.8,1996), e.g., “The I/ CODE Universal Registration
System is an enabling product which benefits both content providers
and Internet users. Sites benefit by obtaining detailed demographic
data while avoiding redundant site-specific registration that negatively
impacts traffic.
*Raw Demographic Data on all I/CODE members who sign-on at your
site is provided free of charge.
* Obtain: data on age distribution, income levels, gender mix, and
other characteristics.
* Gain insights into the depth of repeat visits to your site.
*Access aggregated audience demographics for all I/CODE members
(not just those who register at your site). ·
* Allow visitors to share their anonymous demographics while
respecting their privacy.
* Understand audience preferences and their reaction to your site's
content.”
BURKOWSKI, e.g., Abstract, “The system will provide selective content
delivery based on individual and group profiles, hypertext links into
archival and external data, continuous coverage of news stories,
interactive objects, and "smart" advertising.
BURKOWSKI, e.g., at 2, “Such systems typically provide two types of
218

Reference

Disclosure
services: retrieval of stories (documents) in response to user queries
and personalized clipping services (i.e., selective dissemination of
information) based on user profiles.”
BURKOWSKI, e.g., at 3, “The information content and functionality of
such a system will include
1. Core content: This comprises the stories and advertisements
considered important for all readers. It is transmitted to all users and is
accepted by all clients for display.
2. Stereotyped content: Group profiles or stereotypes can be generated
based on demographic
information linking readers to various sections of the newspaper.
Readers will be categorized by one or more such stereotypes and will
receive various special interest sections, features, advertisements, etc.,
that meet the constraints of these stereotypes.
3. Supplemental content: While reading the news, a reader may
request additional information
by invoking a hypertext link or by querying a multimedia archive.
Such an archive could be
supported directly by the publisher of the newspaper, it could be a
private archive held locally
by the reader, or it could be a distributed archive on the Internet.
4. Individual profiles: The client subsystem will actively gather and
filter information in accordance with an individualized reader profile.
Such a profile might include gender, age, interest areas, income level,
occupation, ethnic background, lists of products in which the reader
has shown an interest, and reading habits such as preferred depth of
news analysis. These last profile attributes will be updated
dynamically as the client monitors the user's reading activity.
5. Advertisements: The system will feature customized, interactive
advertisements that catch
the attention of and involve the reader. These advertisements could
gather information about the
reader so that products and product advertising can be customized and
targeted. This supports the trend to maintain marketing databases that
keep track of customer related information.”
BURKOWSKI, e.g., at 3, “The proposed architecture consists of three
layers, n a distributed client/server environment; the news sources
layer, the news packagers layer, and the
readers layer. The news sources layer consists of news producers that
generate the news items and supply them in some agreed upon markup
format. The news packagers layer consists of client/servers that accept
items from the news sources and produce electronic editions of “the
news”, including advertisements, etc., based on stereotypes. The
219

Reference

Disclosure
readers layer consists of the end-user client/servers. These accept
editions of the news and produce the individual editions of “the news”.
This includes, dynamic layout and assembly and requesting
supplemental material based on the profile or end-user actions. Current
work, not discussed in this paper, is focussing on the details of such an
architecture, scalability, and networking.”
BURKOWSKI, e.g., at 4, “An abstract data representation was defined
and applied to the source data. Using an abstract representation
divorces the display and manipulation of the news items from the
original format of the source. A reader stereotype was defined for the
prototype, as per demographic data supplied by The Chronicle-Herald.
The client selected data from the abstract representation and processed
and formatted it to produce the news display for the reader, based on
the stereotype. The client has control over the display and order of
items in the sections and the order of access to sections, but in this
case, not over content of the sections or the order of the news items in
the sections.”

Tim O’Reilly,
“Publishing Models for
Internet Commerce,”
Vol. 39, No. 6 (1996)
(“O’REILLY”)

NAQVI WO

BURKOWSKI, e.g., at 5, “An extremely important feature of such a
system will be the two-way communications available. Ads will be
able to track who views them, how often, and for how long, and will
be able to report this information to the advertiser. . . In summary, we
feel that the delivery of electronic news is well suited to exploit the
promised high bandwidth, switched, interactive communication
facilities of the information highway. The presentation of such news
will be based initially on a newspaper metaphor and will exploit
communication and multimedia technologies to integrate other news
sources, such as newscasts and video clips, with the text backbone.
The system will provide selective content delivery based on individual
and group profiles, hypertext links into archival and external data,
continuous coverage of news stories.”
O’REILLY, e.g., at 82, “2. There is clear feedback to the advertiser
about what works and what doesn’t, in the form of access logs. This
feature tends to drive advertisers toward providing valuable content
rather than hype. (Unfortunately, many of the people who followed
our lead into net advertising haven’t yet learned that lesson!) In
addition to varying the content of their advertising, advertisers can
experiment with—and get detailed feedback on—the context in which
advertising is most effective. For example, many advertisers are
looking at the click rate---the rate at which readers actually click on an
advertising hyperlink----as well as the overall page views or
impressions in evaluating sites for advertising placements.”
NAQVI WO, p. 3 – “It is a further object of the present invention to
220

Reference

Disclosure
provide a method and system for advertising on a computer
network in which advertisements are more focused and
targeted, for example, by user queries and user profiles,
including the past history of the user's interactions with
the system.”
NAQVI WO, p. 4 – “The present invention provides a new process and
system for online advertising. This new process will be
referred to throughout this application as query-based
advertising ("QBA"). In the QBA process, advertisements
are primarily triggered by user queries. User queries, as
15 used herein, refer to requests from an information consumer
for one or more pages of information from a computer
network. As a result of a query, a user is exposed to
advertisements with the present invention, i.e., the query
triggers advertisements.”
NAQVI WO, p. 5 - “When the user requests a certain page or a certain
topic of information, the relevant pages are retrieved from
the computer network and shown to the user. The present invention,
upon receiving the user's request, retrieves advertisements that are
related to the user's action, dynamically mixes the advertisements with
the content of the pages according to a particular layout, and displays
the pages with focused, targeted advertisements as a part of the page.
The advertisements can be made to satisfy a set of constraints
requested by the advertiser, as well as the constraints of the publisher
of the page, as further discussed below.
The advertisement triggering mechanism of the present
invention is not random or coincidental, but rather, is
prespecified in advance. This specification will be
referred to in this application as a contract. A contract
specifies the marketing rules that link advertisements with
20 specific queries. For example, a diet soft drink
advertisement may be shown when a user asks for a page
about exercising equipment. These rules are specified by
advertisers implementing the concept of "focus" or
"relevance" of advertisements and help the advertisers to
25 target a specific audience. Owners of pages specify the
focus content of their pages through special tags within a
page. These tags are not displayed to the information
consumer; the tags are used to decide what advertisement
can be shown when the page is requested by a consumer.””
NAQVI WO, p. 15-16 – “Initially, a user requests a particular piece of
221

Reference

Disclosure
information through one of the clients 17. The user's
10 request is given to the WWW Daemon 16, which passes the
information to the gate 15. The gate 15 at this point
decides what piece of information is being requested by the
user and finds other relevant pieces of information that
can be commingled with what the user has asked. The user,
15 for example, might ask the system to see certain car
dealers, to find a phone number of a car dealer, or to get
a page of a particular magazine. The gate 15 at this point gives the
request to the matching rule engine 18 ("MRE"). The purpose of the
MRE 18 20 is to look at the content of the user's query and to find a
category within its active index SIC 19 that matches the
same type. If the user has asked for car dealers, the MRE
18 invokes its rules to determine that car dealers are part
of a class of things relating to transportation. Based on
25 the classification determined by the MRE 18, the system now
knows that the user is asking about cars or about
transportation or about whatever else that the user might
be interested in. The MRE 18 at this point then returns to the gate 15
30 the category index of the user's query. If the user had
asked about cars or about family sedans or about sports
cars, at this point the MRE 18 would have figured out that
the user's interest falls into a certain category. Based
on the user's interest category, the system then retrieves
the advertisements that are relevant to that category.
Thus, the purpose of the MRE 18 is to figure out what the
5 user requested, to place the user's request in a category
of a classification system (i.e., the active index SIC 19)
and, based on that classification, to retrieve relevant
advertisements.”
NAQVI WO, p. 20 – “During the computation of the advertisements
and all the other computations that the system of the present
5 invention performs, a logging module 22 of the system
performs extensive logging of what the user has asked, what
advertisements were shown, how long the advertisements were
shown, and which advertisements were shown to which user.
The logging module 22 then stores these logs in a SYS logs
10 database 23. Various scanned reports can be produced and
defined using the information in the SYS logs database 23.”
NAQVI WO, p. 26-27 – “The "focus" arrows 43 shown in Fig. 2
indicate that a certain focus is associated with each category. The
query may have been directed to a category of listings or a particular
vendor. In both cases there is a "focus"
222

Reference

Disclosure
associated with the content of the query (e.g.,
automobiles, physicians, lawyers, etc.). In addition,
there may be a focus associated with the geographic
5 location of the user to permit advertisers to target users
in particular geographic regions. The focus process plays
a major part in the present invention. No advertisements
are shown unless it can be determined that the
advertisements are in some way focused or related to the
10 content of what the user requested.”
NAQVI WO, p. 40 – “The user may also be asked to provide certain
demographic or profile information. For instance, the user
can require that his advertisement be shown only to people
in age group 30 to 40 or only to people living in
Morristown, NJ or any other geographic location. The last
item that the user is asked to specify is the contract.
The various contracts available to the advertiser are
explained above. When the user is finished entering all of
this information, the system updates the ad info database
3 0 ( step 115 ) .”
Figures 1, 2, 7, 10, 11 (and associated text)

BULL

BULL at Col. 3 - “The user is presented with a variety of search,
display and output options. The search options include: 1) Search
using keywords or combinations; 2) Use of complex software text
search agents that have been predefined by the information
aggregation and synthesization system site operators. These agents
take advantage of the expansive subject matter experetise in
understanding which search parameters will best serve the user’s
search needs; 3) Use of search patterns and agents from this user’s
previous sessions, perhaps expanded by available specials and
promotions; 4) Natural Language Query; and 5) Some combination of
1), 2), 3) and 4). During a user session or when a user completes a
session, the user’s looking activity is analyzed for patterns,
preferences and trends and the profile annotated or updated so that
when they next use the information aggregation and synthesization
system, the nominated searches will be customized to their individual
desires.”
BULL at Col. 3 – “The user logs on to the system either by name,
address, etc. or with some pseudoonym (or some combination). This
allows the user’s activity to be tracked and establishes a log of the
user’s activity during the current online experience (session). The user
is also asked for explicit profile information concerning preferences.
223

Reference

Disclosure
These preferences will be used to narrow the information retrieval.”
BULL at Col. 4 - “Along with displays, including those for data entry,
searches, search results, information retrieval, the user will
be presented with advertisements and/or coupons based on
criteria entered by advertisers. This criteria may take the
form of simple logic, linking an ad/coupon with a display or
be derived from complex software text search agents that
analyze one or more of the following: The user’s looking
pattern, the user’s psychographic profile, the user’s personal
profile, the availability of the advertiser’s/couponer’s goods
or services at the instant in time that the criteria is being
exercised. The placement of the ad/coupon will be logged
along with user profile information and provided to the
advertiser/couponer in some form of report.”
BULL at Col. 5 – “IV. Automated Profile Generation.
Presently, user’s profiles are collected based on explicit
entry by the user, and extraction from demographic data
collected from a variety of sources. In the present invention, the
searching patterns of the user on the Internet are monitored. A set of
software text agent profiles is developed and may be integrated with
explicitly collected profile information. The automated profile
generation will have both explicit profile information gathering and
implicit profile information gathering capabilities.
As the user uses the information aggregation and synthesization
system, the pattern of information being viewed is analyzed and the
user presented with search ideas as well as promotions and specials
from suppliers based on these patterns.”
BULL at Col. 6 – “A theme or definition of a class of information (e.g.,
central California travel and tourism or new automobiles) is
identified. Data sources (Local DataStores (500 . . . N) and
Network Accessible DataStores (300 . . . N)) are screened
for relevance, quality of information and appropriateness (or
may be included de facto based on their title or description).
These are indexed using a text indexing software tool 2981
and the indices stored on the system index DataStore 220.
An initial set of Preestablished Software Text Agents are
defined. These agents are words or combinations of words
that form a word based search pattern. This initial set of
agents is relevant to the searches that might be performed
against the class of information that was indexed. (i.e.,
Agents about automobiles would be developed to search a
class of indexed information about new cars). These are
224

Reference

Disclosure
stored in the Preestablished Software Text Agent DataStore
231. The System 200 uses any multipurpose computer
central processing units with the ability to handle multiple
inputs and outputs with the necessary hard disk storage and
to run World Wide Web (WWW) or other network server
software.”
BULL at 7 - “Login and Profiles:
Users using a user access system 100 access the infor
mation aggregation and synthesization system 200 through
the Internet or other public or private network. The user
either logs in by name or by pseudonym or from data
previously stored in the user access system 100. New users
create an account on the user profile datastore 210. Previous
users are identified to an existing account. The user is
presented with a variety of options to create or update profile
information in the user profile datastore 210. This involves
a single data entry option or many mini-options based on the
browsing activity.”
BULL at Col. 7-8 – “The user is also presented with browsing options
based on: activity from a previous session in the browsing activity
datastore 240; predeveloped software text agents and personalized
software text agents (developed in the Post Session Activity) stored in
the Personal Search Text Agent
DataStore 232; or combinations of all as well as situational
opportunities developed by the user greeting subsystem 291.
The user selects the search options to be used (or simply
enters search criteria directly). This search criteria is used to
search the index datastore 220 and a list of data sources is
presented to the user for selection. The user indicates the
information to be viewed. The user will also be presented
with options to refine his search through the altering of
search agent criteria (Search Reduction System 293).”
BULL at Col. 10 - “User Profile DataStore
This contains data about the user, preferences, situational
preferences, accounting information, psychographic profile,
personal profile and other relevant information related to the
user by individual identifier.”
BULL at Col. 10 – “232 Personal Search Text Agents
These are complex software text search patterns that may
be individual words or word sets and/or combinations of
words and Preestablished Software Text Agents 231 includ
225

Reference

Disclosure
ing the results of the post session analysis 2921 that provide
individually customized searching of the Index DataStore
220.
BULL at Col. 12 – “IV. Automated Profile Generation
Browsing patterns of the user are analyzed and these
patterns update profiles automatically.
FIG. 7 illustrates a how diagram for the Automated Profile
Generation. The looking patterns of the user are monitored to develop
a set of software text agent profiles that are integrated with explicitly
collected profile information to assist the user in
narrowing down information for future sessions as well as
suggesting references, merchandise or services during the
current session. This is accomplished by statistical analysis
of the text stream.
The searching patterns of the user on the Internet are
monitored by monitoring the text stream. A set of software
text agent profiles is developed and may be integrated with
explicitly collected profile information. The explicit infor
mation is gathered by queries to the user. The explicit and
implicit data are merged to develop software text agents that
support the user’s future shopping sessions.”
BULL at Col. 12 – “Certain criteria will be entered which delineates a
pattern that is requested to be monitored. When this pattern is seen (or
is in close match) in the user’s WWW activity, the insertion
mechanism is activated. If a certain web page is
requested, the present invention will display a particular
advertisement. The ad will be inserted based on the content
of the existing web page being read. An analysis of the text
stream of the user’s interactive session will be performed
online. When certain text patterns are observed (or close
matches are observed), an advertisement is inserted into the
display. The advertising may be static or connected to the adver
tiser’s computer datastore which designates specific ads or
coupons based on the pattern match and other conditions
which may be required. The software agent criteria is entered by the
merchant in the agent data store 230 which delineates a pattern that
needs to be monitored.
As an example, if the user accesses web pages for
“Holiday Inns on the West Coast”, the insertion mechanism
Would be established to automatically insert ads for “Hilton
Inns on the West Coast.””
BULL at Figs. 1 - 7 (and associated text)
226

Reference

Disclosure

KOHDA ’96

KOHDA ’96, §1: “An advertising agent is placed between the
advertisers and the users. Advertisements fetched from advertisers'
Web servers are merged with Web pages from ordinary Web servers
by the agent, and the merged pages are displayed on the users' Web
browser. Thus, the users see advertisements on any server around on
the Internet. Moreover the agent has chances to deliver appropriate
advertisements which suit each user's taste.”
Id., §2.2: “Note that the agent is aware of the identity of the user and
which page the user is about to read on the browser, so the advertising
agent can tailor advertisements for individuals and their current
interests. Thus it prevents the user from having to see advertisements
that are unrelated to their current interests.”
Id., §3.1: “At invocation, environment information is passed to each
filter program as invocation parameters. The environment information
includes at least the identity of the user and information about the
selected anchor. The contents of a Web page designated by the anchor
are input into the pipe of filters, and the output from the pipe is
displayed on the browser's window as an HTML document.”
Id., §3.2: “The filter keeps in memory the contact path (URL) to the
agent's Web server. When it is invoked, it forwards the invocation
parameters passed from the browser to the agent's Web server, and
waits for a reply.”

KOHDA ’853

KOHDA ’853 at 6:56 to 7:3: “The user inputs data for use in obtaining
requested retrieved information (for example, articles from a
newspaper relating to a specified item) through the input/output unit 1.
Then, the information retrieving apparatus 100 obtains the retrieved
information from the information retrieving server through the
retrieved information obtaining unit 3, automatically obtains additional
information such as advertising information from the information
server through the additional information obtaining unit 4,
incorporates the obtained information into the retrieved information
obtained from the information converting unit 2, and outputs the result
on a display unit.”
Id. at 7:32-43: “The information retrieving apparatus 100 can be
widely applied to, for example, advertisements through the WWW.
That is, sufficient advertising effect can be gained even when access is
concentrated on a very small number of popular information servers,
and a large number of other information servers are rarely accessed.
227

Reference

Disclosure
Therefore, a sufficient number of advertisers can be collected. In the
WWW, information users are individuals and there are not a large
number of users concentrated in one access operation. However, since
the advertising information has been preliminarily selected by the user,
the user is interested in the provided advertisement in most cases.”
Id. at 14:16-54: “The additional information use history storage unit
51 stores an actual use history of the user corresponding to the
additional information. That is, the additional information use history
storage unit 51 stores a private history in its memory if the data
required to obtain retrieved information from the retrieval condition
input unit 11 is in the additional information…. The frequency of uses
refers to the number of times the information is used. Especially if the
additional information is advertising information, it is also recorded
whether the product in the advertisement has been purchased through
the additional information. … For example, when the additional
information describes a new personal computer of a specific
manufacturer, it can be obtained as the detailed information about the
practical specification, appearance, etc. of a desired model. The
information is instructive for the user, and is also useful for the
advertiser because it improves an advertising effect for the product.”
Id. at 18:64 to 19:4: “The use of such additional information is
recorded in the additional information use history storage unit 51. For
example, the number of times the information is used is recorded ‘+2’
because the detailed information is obtained from the advertisement,
and the contract information is obtained from the detailed information.
When a purchase contract is signed for the advertised product, it is
also recorded.”
Id. at 15:65 to 16:2: “the information server 102 or advertising agent
server 102A reads the additional information use history at
predetermined intervals to be informed of the tendency of liking of the
user.”

“Firefly Licenses
Targeting Technology,”
by Debra Ahe
Williamson, December
9, 1996, available at
adage.com/article/news
/firefly-licensestargetingtechnology/75969.
(“WILLIAMSON”)

See e.g., WILLIAMSON, p. 1 (identifying Yahoo! as licensing Firefly’s
technology; “Firefly users provide basic demographic information,
such as age, gender, ZIP code and e-mail addresses. As they traverse a
site, entering different content areas and rating their interests, that
information is added to a user profile.”); id. (“Participating sites will
use Firefly’s Passport software to register visitors and build individual
profiles based on visitors’ activity on a site.”)

228

Reference
“Firefly Network and
Yahoo! Offer
Consumers Ability to
Intelligently Navigate
the Web; My Yahoo!
Features Firefly Tools
to Offer Personalized
Recommendations for
Web Sites and Build
Dynamic
Communities,” Dec.
11, 1996 (“FIREFLY
NETWORK AND YAHOO!
OFFER CONSUMERS
ABILITY TO
INTELLIGENTLY
NAVIGATE THE WEB”)
“Boston.Comment
Today’s topic Shadow
advertising,” The
Boston Globe,
November 14, 1996.
(“BOSTON GLOBE”)
ABOUT NETGRAVITY
ADSERVER

Lang, “NewsWeeder:
Learning to Filter
Netnews,” 1995
(“LANG”)

Green, Bayer &
Edwards, “Towards
Practical Interface

Disclosure
See e.g., FIREFLY NETWORK AND YAHOO! OFFER CONSUMERS ABILITY
TO INTELLIGENTLY NAVIGATE THE WEB, p. 1 (“Using Firefly software
tools, customer sites can register and recognize Firefly PassportTM
holders, deliver personalized recommendations, create relevant and
dynamic communities, serve targeted content and ads and more
accurately measure and report on site activity.”); id., p. 2: “The
Passport Office also enables Firefly software tools customers to
deliver targeted content and advertising, as well as, accurate
measurements and reports regarding site activity.”

See e.g., BOSTON GLOBE, p. 1 (“Firefly offers advertisers, a movie
studio, for example, the opportunity to deliver an ad plugging a new
Bruce Willis movie only to users who have rated previous Bruce
Willis movies highly.”)

See e.g., ABOUT NETGRAVITY ADSERVER, Getting Started, p. 1
(“AdServer uses a sophisticated scheduling algorithm to select the ad
to show, reading the ad and scheduling information from its database.
AdServer evaluates many scheduling criteria for choosing an ad,
including . . . user profile targeting.”); id., NGAPI Basics, p. 1 (“Such
custom functions may perform the following actions: target ads to
users based on browser cookie information or lookups in a custom
database”
See e.g., LANG, Introduction (“the user can also use NewsWeeder’s
virtual newsgroups. For example, user Bob might go to the virtual
newsgroup nw.top50.bob to see NewWeeder’s personalized list of the
top 50 out of all articles, according to learned preferences for Bob. He
is then presented with a list of one-line article summaries, sorted by
predicted rating. The user selects a group of articles from these
summaries and reads them sequentially. After each article is read, the
user clicks on a rating from one to five . . . NewsWeeder collects the
user’s ratings for active feedback on the user’s interests. . . . Each
night, the system uses the collected rating information to learn a new
model of the user’s interests.”)
See e.g., GREEN, Introduction (“The agent is given a minimum of
background knowledge, and learns appropriate behavior from the user
and perhaps other agents. The use of machine learning methods to
229

Reference
Agents which Manage
Internet-Based
Information, 1995
(“GREEN”)

MEEKER

Disclosure
develop a profile of user preferences allows the agent to adapt to
changes in user behavior, as well as eliminating the need for explicit
programming with rules or scripts. A common method of developing
a user profile is by observing and analyzing user behavior.”); id. (“The
user profile is then employed to generate classifications for the new
documents, such as a user’s interest rating in a USENET news article
or a World-Wide Web page.”); id., Section 4 (“LAW helps a user find
new and interesting information on the World-Wide Web. It provides
assistance in two ways: by interactively suggesting links to the user as
they browse the Web and through the use of a separate Web robot that
attempts to find pages that might be of interest.”)
MEEKER at v.: “However, that same marketer should get even more
interested if a Web site (such as CNET, at www.cnet.com)
can route advertisements to a demographic group that includes only
males who are at least 35 years old, have household incomes in excess
of $100,000, live in California, and use Pentium PCs with Netscape
Navigator.”
Id. at 3-13: “However, for direct marketing, the Internet offers the
ability to target and deliver messages to an audience with specific
demographics and interests, and allows the user to interact instantly
with that message. In essence, direct response advertisers sell goods
and services to customers individually, and no other medium affords
users such immediate access at the point of sale.”
Id. at 6-2: “Each time the page is downloaded by a user, a designated
space on the page (in the example in Figure 6-1, a rectangle across the
top) is automatically filled with a banner. The method by which a site
determines which ad to put into which download may depend on
agreements or contracts with advertisers, the capability of the
technology involved, the demographics of the user, and other factors.”
Id. at 6-3: “This brings us to the concepts of inventory management
and allocation, and ad tracking and rotation. The most important goal
of advertising is to deliver to each person the message most
appropriate to their tastes, buying habits, and so forth, and with the
most effective frequency — in other words, to execute a campaign
tailored to each individual. To this end, many Web sites use software
packages to impose ad delivery schema over on-the-fly allocation of
advertising inventory. By schema, we mean sets of rules governing
which ads get delivered when. This software can be either off-the-shelf
(from companies like Net Gravity, Bellcore, and Accipiter) or
developed in-house (as HotWired and CNET, for instance, have done).
The importance of the quality, flexibility, and reliability of ad
management software is simple: more targeted, reliable, and verifiable
230

Reference

Disclosure
advertising delivery translates directly into the ability to charge more
per impression…. Targeting gives advertisers the opportunity to filter
messages to selected audiences based on certain criteria. This
may be the most powerful aspect of the Internet as an advertising
medium — the ability to dictate the exact composition of an
advertisement’s audience…. This targeting ability has two pieces: 1)
the process for ad delivery and measurement is precise and directed
(e.g., each ad is individually delivered in response to a user-generated
request — there is no TV- or radio-like “shotgun” delivery —
followed by statistical sampling and averaging to determine the actual
composition of the receiving audience); and 2) each individual
delivery can be tailored, based on user information. The power of the
second aspect is increased substantially with more detailed user data,
potentially collected through registration or in the course of using the
site. Thus, with the right user information, one could know that every
advertisement delivered is received by teenage women using a
Macintosh, or by college-educated middle-age men in specific
(perhaps high-income) zip codes, and so on. Essentially, it’s a
marketer’s dream.”
Id. at 6-6: “Search engines, by definition, use text input by users to
conduct searches of relevant content on the Web. Since advertisements
are displayed along with the search results, these companies allow
advertisers to buy “key words,” which display the advertiser’s banner
when a user searches for the word purchased. It follows that the word
or words purchased are generally related in some way to the
advertiser’s products or services. Infoseek and Yahoo! charge $1,000
per month per keyword, and based on a target of 20,000 impressions,
this would yield a CPM of $50. For example, Figure 6-3 shows how
the results of a search for the word “router” yielded a typical list of
sites but also netted an advertisement for Cabletron Systems (a maker
of switches, considered an alternative to routers). In fact, any time this
word was searched for, the same ad came up. A search for “hub”
consistently resulted in a different ad for the same company. (Yes, we
searched for “beer,” and each time we got a Miller Genuine Draft ad).”
Id. at 6-10: “We reiterate our belief that the ability to marry content to
creative will be a key driver of pricing. Essentially, this requires that
the advertising be targeted at the audience for the particular site’s type
of content. The next logical step in this process would then be to tailor
not just to the audience, but also to each individual user according to
his or her buying and browsing habits. Several makers of
personalization software, most notably Firefly Network (formerly
Agents, Inc.), provide products that personalize ad delivery based on a
user’s past behavior or profile. If a user has come to an advertiser’s
231

Reference

Disclosure
site three times, looked at the same item each time, but has yet to
purchase it, delivering an ad for that product as the user again enters
the site would certainly be more valuable to an advertiser than the
delivery of that ad indiscriminately. Once again, the more targeted the
audience delivered, the higher the price advertisers will pay.”
Id. at 7-11: “Another development in this area is the use of cookies,
wherein a server-specific file is sent by a Web site server and
automatically stored by a browser on a user’s hard disk. This cookie
file’s data can be anything, like a date/time stamp, an IP address, or a
unique user ID. Once a cookie is received from a given server,
whenever that browser makes a request to that server for an HTML
page, it will include the cookie with the request. The browser will only
send a cookie to the server site that originally sent it, so it is not
possible for one Web site to look at or request cookies from other
sites. Cookies provide a signature, so that Web sites can track an
individual’s number of visits and the path he or she took through a
site. This information can be employed in a number of creative ways,
including obtaining behavioral data, crafting marketing messages for a
site owner’s or advertiser’s products, keeping track of purchasing
activity at a site (if you visit and read all of my pages on espresso
makers, but don’t buy one, I can still show you the product each time
you return), and overall personalization of the user’s experience at the
site. Some potential downsides to the use of this technology is the
possibility of tampering by users or third parties. Cookies are located
on a user’s local hard drive, and if altering the cookie data is beneficial
enough to a user, it is likely that many will attempt to do so. In
addition, third-party sites might have cause to tamper with the cookie
data of competitors (or partners), or invade the privacy of users by
reading their stored data for behavioral, purchasing, or other purposes.
Despite these potential security and privacy issues, this tailored
marketing approach adds significant value, we believe, that may be
enhanced further by demographic information gained through user
registration data, which are collected at such sites as CNET, ESPNET
SportsZone, The Wall Street Journal Interactive, and the online
services. In our view, it would make a very compelling value-added
proposition if advertisers could be certain of the age, gender,
occupation, or purchasing preferences of each person who views an
ad.”
Id. at 10-10: “Firefly Network (formerly Agents, Inc.; Cambridge,
Mass.; www.firefly.com) was founded in 1995 and provides software
that uses advanced algorithms based on certain collaborative filtering
technologies to make recommendations to users based on their
preferences. In January 1996, the company (then called Agents, Inc.)
232

Reference

Disclosure
launched this intelligent agent technology on the Web in the form of
Firefly. As a user continues to visit the network, Firefly's technology
“learns” his or her likes and dislikes, can compare and contrast these
with other users’ patterns, and is able to offer members personalized
recommendations for music, movies, and so forth. This technology
therefore offers marketers the ability to target messages and
advertisements based on an individual’s preferences and interests. As a
result, marketers can maximize efforts on a prequalified audience and
offer a more relevant experience for consumers. The company
currently has 95 employees and more than 500,000 registered
members. Firefly Network’s customers and partners include: Yahoo!,
Ziff Davis’s ZD Net, Reuters, Rolling Stone, Newbury Comics, The
All Music Guide, Hits World, and Muzak’s Enso Audio Imaging
Division. They have raised in excess of $18 million from investors,
including: Atlas Ventures, Dun & Bradstreet Enterprises, Merrill
Lynch, PAFET, Softbank, Trident Capital, Goldman Sachs, and
Reuters New Media.”

U.S. Patents No.
6,183,366 to Goldberg
et al. (“’366 PATENT”)

’366 PATENT, e.g., Abstract, “The present invention is an information
service and advertising providing system for presenting interactive
information services together with interactive advertising on a
communications network such as the Internet and LANs. The
information service may be a game played interactively on the
network while advertising is communicated between users and an
advertising network node. However, other interactive services, such as
are available on the Internet, are also accessible for concurrent use
with advertising presentations. Advertising or promotionals may be
selectively presented to users by comparing archived user profiles with
demographic profiles of desired users. User responses to advertising
may be used for evaluating advertising effectiveness such as for test or
microtarget marketing. Compensation to users for viewing advertising
may also be provided. For instance, users may be provided with
subsidized Internet access for receiving advertising while concurrently
interacting with an Internet service. Users may also be provided with
various games and/or game tournaments via interactive network
communications. Thus, users may respond to advertising while being
entertained (e.g., via games), or while interacting with another network
service.”
’366 PATENT, e.g., Summary of the Invention, “The present invention
is a computerized interactive advertising system (i.e., method and
apparatus) for exchanging information regarding goods and/or services
between a first population of users (hereinafter also known as
"players" or "users") and a second population of users (hereinafter also
233

Reference

Disclosure
known as "sponsors" or "advertisers"). In particular, the sponsors or
advertisers may present information related to goods and/or services to
the players using the present invention and the players may view this
information while, for example, interacting with the present invention
for playing a game such as blackjack, craps, roulette, poker, pai gow
or the like. Moreover, a player may also interact with the present
invention so that the player has the capability for responding to
sponsor or advertiser presented questionnaires, as well as for
purchasing or viewing sponsor goods and/or services. Thus, the
present invention provides an information exchange service within a
gaming context for enticing players to view and/or interact with
sponsor presentations such as interactive advertisements.
It is also an aspect of the present invention that each player or user is
presented with advertisements for products and/or services, wherein it
is believed the player will be receptive to the advertisement. That is,
the present invention selectively presents advertisements to each
player, according to stored characteristics and preferences of the
player that the present invention has determined from, for example,
player supplied personal information, player responses to questions,
and/or analysis of player interactions such as player requests for
additional information related an advertisement. Thus, such a selective
presentation of advertisements allows a sponsor or advertiser to
provide information related to relatively extensive or expensive
promotionals (e.g., demonstrations, samples, discounts, trial
subscriptions, prizes, bonuses) to players most likely to subsequently
purchase the advertised product or service. Consequently, such
selectivity can greatly increase the cost effectiveness of advertising,
wherein the term, advertising (or advertising presentation), as used
herein is understood to include not only product or service
presentations that are merely informational, but also more interactive
advertising presentations such as promotionals wherein discounts, free
samples or a trial usage may be offered. . . It is a further aspect of the
present invention to require each player to use a distinct identification
provided when the player "registers" with the present invention before
playing any games so that a network site for the invention may be able
to identify each player. Accordingly, it is an aspect of the present
invention during registration, that each player provides personal
information about him/herself both for gaming identification and/or
use as selection criteria by sponsors or advertisers for presenting
particular presentations. For example, in the case of an Internet
embodiment of the present invention, such registering can be
performed via the Internet prior to play of any games at a
gaming/advertising web site. Thus, players may be required to provide
the present invention with information about themselves such as name,
address, E-mail address, age, sex, and/or other player characteristics
234

Reference

Disclosure
deemed pertinent to one or more sponsors or advertisers. Accordingly,
the present invention provides a sponsor or advertiser with the
capability to target its presentations substantially only to players or
users having selected characteristics as, for example, determined from
player information provided when registering with a network site for
the present invention.”

’366 PATENT, e.g., Claims 1, 2, 3
’366 PATENT, e.g., Figures 3, 4A-E, 6A-B, 7 (and associated text)
U.S. Patents No.
7,496,943 to Goldberg
et al. (“’943 PATENT”)

’943 PATENT, e.g., Abstract, “A networked system is disclosed for
presenting advertising during on-line interactions between a user and a
service of a network (e.g., the Internet, interactive cable, and/or a
LAN). Advertisements (ads) are presented to a networked user
unrequestedly during user interactions with the service. The user can
activate the ads (via hyperlinks) for receiving additional advertising.
The system gathers user data and/or develops user profiles for
selectively presenting ads, promotionals, discounts, etc. targeted to
receptive users. In exchange for viewing such selective presentations,
on-line access to the service is provided, the service including, e.g., (a)
playing on-line interactive games (e.g., blackjack and poker), (b)
providing access to the network itself (e.g., an Internet service
provider), and/or (c) providing access to substantially any interactive
service accessible via (b). The system can provide free/reduced cost
network services to the user for viewing unrequested advertising. The
system can be provided for a casino.”
PATENT, e.g., Summary of the Invention, “he present invention is a
computerized interactive advertising system (i.e., method and
apparatus) for exchanging information regarding goods and/or services
between a first population of users (hereinafter also known as
“players” or “users”) and a second population of users (hereinafter
also known as “sponsors” or “advertisers”). In particular, the sponsors
or advertisers may present information related to goods and/or services
to the players using the present invention and the players may view
this information while, for example, interacting with the present
invention for playing a game such as blackjack, craps, roulette, poker,
pai gow or the like. Moreover, a player may also interact with the
present invention so that the player has the capability for responding to
sponsor or advertiser presented questionnaires, as well as for
purchasing or viewing sponsor goods and/or services. Thus, the
present invention provides an information exchange service within a
gaming context for enticing players to view and/or interact with
235

Reference

Disclosure
sponsor presentations such as interactive advertisements.
It is also an aspect of the present invention that each player or user is
presented with advertisements for products and/or services, wherein it
is believed the player will be receptive to the advertisement. That is,
the present invention selectively presents advertisements to each
player, according to stored characteristics and preferences of the
player that the present invention has determined from, for example,
player supplied personal information, player responses to questions,
and/or analysis of player interactions such as player requests for
additional information related an advertisement. Thus, such a selective
presentation of advertisements allows a sponsor or advertiser to
provide information related to relatively extensive or expensive
promotionals (e.g., demonstrations, samples, discounts, trial
subscriptions, prizes, bonuses) to players most likely to subsequently
purchase the advertised product or service. Consequently, such
selectivity can greatly increase the cost effectiveness of advertising,
wherein the term, advertising (or advertising presentation), as used
herein is understood to include not only product or service
presentations that are merely informational, but also more interactive
advertising presentations such as promotionals wherein discounts, free
samples or a trial usage may be offered.
Moreover, it is an aspect of the present invention that each player may
interact with and play a game at a time and pace (i.e., tempo)
substantially of the player's choosing. In particular, the player is not
bound by a required order or sequence of play involving other players,
even though the player may be in competition with other players. In
fact, a player may cease play for an extended time while in the midst
of a game and subsequently continue the game at the point where the
player ceased to play. Thus, if the present invention is easily
accessible, then players may interact with the present invention at their
leisure. . . “t is a further aspect of the present invention to require each
player to use a distinct identification provided when the player
“registers” with the present invention before playing any games so that
a network site for the invention may be able to identify each player.
Accordingly, it is an aspect of the present invention during
registration, that each player provides personal information about
him/herself both for gaming identification and for use as selection
criteria by sponsors or advertisers for presenting particular
presentations. For example, in the case of an Internet embodiment of
the present invention, such registering can be performed via the
Internet prior to play of any games at a gaming/advertising web site.
Thus, players may be required to provide the present invention with
information about themselves such as name, address, E-mail address,
age, sex, and/or other player characteristics deemed pertinent to one or
more sponsors or advertisers. Accordingly, the present invention
236

Reference

Disclosure
provides a sponsor or advertiser with the capability to target its
presentations substantially only to players or users having selected
characteristics as, for example, determined from player information
provided when registering with a network site for the present
invention.”
’943 PATENT, e.g., Claims 1, 2, 9, 11-14
‘943 PATENT, e.g., Figures 3, 4A-E, 6A-B, 7 (and associated text)

U.S. Patents No.
6,712,702 to Goldberg
et al. (“’702 PATENT”)

’702 PATENT, e.g., Abstract, “The present invention is a game playing
method and apparatus for automating games such as blackjack, poker,
craps, roulette, baccarat and pai gow, wherein players may play
continuously and asynchronously, and information related to
advertised items can be exchanged between players and advertisers. In
one embodiment, each instance of a game is likely unique from all
other current game instances. The games do not require a manual
dealer and in one embodiment, played in a gaming establishment using
low cost gaming stations. The present invention may also, be used to
play such games on the Internet or an interactive cable television
network wherein a game controller communicates with players at
network nodes in their homes and at their leisure since there is no
game tempo requirement. During a game, advertising is selectively
provided by comparing player personal information with a desired
demographic profile. Player responses to advertising are used for
evaluating advertising effectiveness. The invention is useful for test
marketing of products, advertisements, and reduces advertising costs.”
‘702 PATENT, e.g., Summary of the Invention, “he present invention is
a computerized interactive advertising system (i.e., method and
apparatus) for exchanging information regarding goods and/or services
between a first population of users (hereinafter also known as
“players” or “users”) and a second population of users (hereinafter
also known as “sponsors” or “advertisers”). In particular, the sponsors
or advertisers may present information related to goods and/or services
to the players using the present invention and the players may view
this information while, for example, interacting with the present
invention for playing a game such as blackjack, craps, roulette, poker,
pai gow or the like. Moreover, a player may also interact with the
present invention so that the player has the capability for responding to
sponsor or advertiser presented questionnaires, as well as for
purchasing or viewing sponsor goods and/or services. Thus, the
present invention provides an information exchange service within a
gaming context for enticing players to view and/or interact with
237

Reference

Disclosure
sponsor presentations such as interactive advertisements.
It is also an aspect of the present invention that each player or user is
presented with advertisements for products and/or services, wherein it
is believed the player will be receptive to the advertisement. That is,
the present invention selectively presents advertisements to each
player, according to stored characteristics and preferences of the
player that the present invention has determined from, for example,
player supplied personal information, player responses to questions,
and/or analysis of player interactions such as player requests for
additional information related an advertisement. Thus, such a selective
presentation of advertisements allows a sponsor or advertiser to
provide information related to relatively extensive or expensive
promotionals (e.g., demonstrations, samples, discounts, trial
subscriptions, prizes, bonuses) to players most likely to subsequently
purchase the advertised product or service. Consequently, such
selectivity can greatly increase the cost effectiveness of advertising,
wherein the term, advertising (or advertising presentation), as used
herein is understood to include not only product or service
presentations that are merely informational, but also more interactive
advertising presentations such as promotionals wherein discounts, free
samples or a trial usage may be offered.
Moreover, it is an aspect of the present invention that each player may
interact with and play a game at a time and pace (i.e., tempo)
substantially of the player's choosing. In particular, the player is not
bound by a required order or sequence of play involving other players,
even though the player may be in competition with other players. In
fact, a player may cease play for an extended time while in the midst
of a game and subsequently continue the game at the point where the
player ceased to play. Thus, if the present invention is easily
accessible, then players may interact with the present invention at their
leisure. . . “It is a further aspect of the present invention to require each
player to use a distinct identification provided when the player
“registers” with the present invention before playing any games so that
a network site for the invention may be able to identify each player.
Accordingly, it is an aspect of the present invention during
registration, that each player provides personal information about
him/herself both for gaming identification and for use as selection
criteria by sponsors or advertisers for presenting particular
presentations. For example, in the case of an Internet embodiment of
the present invention, such registering can be performed via the
Internet prior to play of any games at a gaming/advertising web site.
Thus, players may be required to provide the present invention with
information about themselves such as name, address, E-mail address,
age, sex, and/or other player characteristics deemed pertinent to one or
more sponsors or advertisers. Accordingly, the present invention
238

Reference

Disclosure
provides a sponsor or advertiser with the capability to target its
presentations substantially only to players or users having selected
characteristics as, for example, determined from player information
provided when registering with a network site for the present
invention.”

’702 PATENT, e.g., Claims 1, 3, 4, 12
‘702 PATENT, e.g., Figures 3, 4A-E, 6A-B, 7 (and associated text)

PHILLIPS BUSINESS

DEDRICK 1994

PHILLIPS BUSINESS at 1: “But most vendors also have more to offer
than just high volume, thanks to such approaches as "narrow casting,"
or placing ads based on key words entered in a search. These
capabilities allow advertisers to target audiences through search
engines like no other medium. "Not only can the engines track the
things you're searching on, they can suggest target ads. This is one-toone marketing," Julian said.”
PHILLIPS BUSINESS at 1: “While search engines can personalize ads
based on search terms, another effective model is to personalize entire
sections based on geographic and demographic factors. Vendors can
not only index content for a targeted population, they can sell
advertisers a guaranteed demographic.”
See e.g., DEDRICK 1994, p. 57: “Consumer demographic and
psychographic data are important to advertisers, because these are the
data that allow an advertiser to target specific consumers.
Demographic data include variables such as age, sex, income, marital
status. Psychographic data include likes and dislikes, color
preferences and personality traits that show consumer behavioral
characteristics. The better the demographic and psychographic data
available on a set of consumers, the better an advertiser is able to
target an advertisement to this set of consumers.”); id., p. 59: “dad, a
male, age 40-50, earning $70,000+ annually, may be presented with a
portion of a Mt. FunSki ski resort advertisement concerning booking a
reservation, along with a list of fun things to do. However, the
consumer’s son, male, age 12-17, interested in girls, moguls, and hot
tubs, may consume a presentation based totally upon on [sic] the ‘fun
things’ that Mt. FunSki has to offer.”); id. (“consumers will have
personal profiles residing within their consumption devices. These
personal profiles may contain demographic and psychographic
variables as well as other data. Such data may included a preferred
payment method (Visa, Amex, etc. card numbers) enabling consumers
to easily participate in electronic commerce. Other included data
239

Reference

DEDRICK 1995

Disclosure
might include key words and other variables used by consumption
agents to go out on the network and find both electronic content and
electronic advertisements that have a certain “hit-rate” when matched
against a consumer’s profile. Additionally, the consumption device
may have resident software that monitors consumption behavior on an
ongoing basis, allowing a consumer’s personal profile to be
automatically build and maintained. . . . they may begin to see
advertisements that focus on their favorite subjects, presented
primarily in their favorite colors. Also, consumer’s agents may report
the availability of electronic content and advertisements matching their
personal profiles.”); id. (“Acting upon the consumer’s personal profile
data, an agent might send out queries to electronic yellow pages
service providers, either locally or with a wider scope of interest.”);
(“consumer’s agents may report the availability of electronic content
and advertisements matching their personal profiles.”); id., p. 60
(“Additionally, the consumption device may have resident software
that monitors consumption behavior on an ongoing basis, allowing a
consumer’s personal profile to be automatically build and maintained.
. . . they may begin to see advertisements that focus on their favorite
subjects, presented primarily in their favorite colors. Also, consumer’s
agents may report the availability of electronic content and
advertisements matching their personal profiles.”); id., p. 60 (“More
advanced agents may be given access to a consumer’s credit
information and authority to use the credit information, enabling the
agent to conduct electronic commerce on behalf of the consumer.”
); id., p. 62-63 (“the currently suggested attribute extension list is as
follows: . . . Access control attributes, to limit access to electronic
advertisements not available to all consumers, such as advertisements
for alcohol, tobacco, and adult products, . . . Scope attributes,
describing global, national, regional, and local preferences for
distribution and announcement to yellow page services, Language
support attributes, detailing which languages are supported by each
object and providing network pointers to parallel objects authored
using different languages . . .”); id., p. 63 (“Consumer’s personal
profiles may include such variables as a collection of the consumer’s
consumption characteristics, a collection of demographic and
psychographic variables, bank account and credit card account
numbers.” )
See e.g., DEDRICK 1995, p. 43 (“As another example of attribute
extensibility, an element made available to a consumer could depend
on that particular consumer’s target characteristics. For example, Dad,
a male age 40 to 50, earning $70,000-plus annually, might see part of a
Mt. FunSki ski resort ad about booking a reservation . . . However, the
consumer’s son, male, ate 12 to 17, interested in girls, moguls, and hot
tubs, might see a presentation based totally on the ‘fun things’ that Mt.
240

Reference

GALLAGHER

Disclosure
FunSki has to offer. . . .”); id., p. 45 (“consumers will have portable
personal profiles tied into their consumption devices. These portable
profiles may contain such data as a preferred payment method (credit
card numbers, for example) enabling consumers to easily participate in
electronic commerce. Other profile data might include key words and
other variables used by consumption agents for finding both electronic
content and electronic ads that have a certain ‘hit rate’ when matched
against a consumer’s profile.”); id., p. 45 (“a manual profile
modification program is also required to enter personal data such as
name, address, telephone numbers, credit card and bank account
numbers, and the like . . . The consumer can use the manual profile
modification program to correct such deviances from the actual
electronic content consumption preferences.”); id., p. 45 (“2. When a
consumption device presents one of these labeled electronic ads to a
consumer, all input and output between the consumer and the
multimedia element currently being consumed is monitored. 3. Each
of these I\O interactions is correlated to the labels associated with the
particular multi-media element being displayed on the consumption
device. 4. Relations between the elements of the electronic ad that are
not chosen for interaction by the consumer are also correlated with the
labels associated with each multimedia element. 5. The correlations
made in the previous steps are entered into the consumer’s profile,
representing data on what a consumer likes and dislikes.”); id., p. 46
(“Acting upon the consumer’s personal profile data, an agent might
send out queries to electronic yellow pages service providers, either
locally or with a wider scope of interest.”); id., p. 46 (“As personal
consumption profiles become more robust, consumers might begin to
see ads focusing on their favorite subjects, presented primarily in their
favorite colors, sizes and shapes. Also, their agents might report the
availability of electronic content and ads matching their personal
profiles.”); id., p. 46 (“If a cost or rebate is attached to each available
element, the agents could report the monetary units involved with
potential consumption. The agent could leave the final busy/sell
decision up to the consumer, or perform the transaction if programmed
to act on the consumer’s behalf.”)
See e.g., GALLAGHER, p. 3 (“Current technology provides the
capability to develop sophisticated and detailed profiles of individual
users of information services based on individual characteristics and
past patterns of behavior in using the information service.”); id. (“ It is
possible to target users very precisely because data can remain
associated with individuals, so advertisers can select exactly the users
to whom they wish their advertising to be exposed.”); id., p. 4 (“the
model requires that users be assigned unique identifiers . . . Users also
complete an online questionnaire the first time they use the
information service. The questionnaire allows data to be collected on
241

Reference

NETGRAVITY
ADSERVER CHOSEN BY

Disclosure
several dimensions, including: (1) demographic attributes such as
geographic location, income, family lifecycle stage, occupation, and
sex; (2) psychographic attributes such as travel patterns and hobbies;
and (3) product and brand usage attributes. This element of the basic
model permits a banner advertisement to be directed to users (and only
those users) who fit certain criteria, assuming data were collected on
relevant attributes.”); id. (“Each time a user connects, his/her profile is
compared to all target audience profiles from all advertisers. The
user’s profile will actually match some subset of those profiles.”); id.
(“In summary, the basic model has three elements: individual user
profiles, individual advertisement target audience profiles, and a
selection mechanism for presenting advertisements to specific users
who match the target audience profile.”); id., p. 5 (“In the enhanced
model, we propose that patterns of search and browsing behavior
exhibited by users while using an information service determine which
advertisements are shown to that user during current or future
sessions.”); id., p. 5 (“As before, this model relies on assigning a
unique identifier to each user for recording his/her searching and
browsing activities while using the information service. Each session
constitutes a ‘record,’ consisting of data such as: sites visited in order;
pattern of navigation through a hierarchical structure (as in Yahoo!);
choice of search terms in keyword-based searches; and reaction to
previously exposed targeted banner advertisements (e.g., which linked
Web sites are selected and visited by the user and which ones are
ignored). The aggregate of such records for each user provides a
profile from which preferences can be implicitly generated. As a
simple example, if a user has made several searches using keywords
such as ‘Atlantic salmon’ and ‘fly fishing,’ and has visited the site of
the Angling Club Lax-a of Iceland . . ., s/he may be targeted for a
banner advertisement for a fishing lodge in Alaska.”); id., p. 7
(“Profiles accommodate the possibility that some users within the
region of acceptability may be more desirable to an advertiser than
others. Hen, a distance metric capturing the relative desirability of a
user with respect to an ideal profile is possible. . . . recognizing a
notion of distance allows the possibility for advertisers to ‘bid’ for the
opportunity to display an advertisement to a user. Such bids would be
determined by the advertiser, based on variables such as the user
profile . . . and advertising budget.”); id., p. 8 (“When bids are
received, they can be ranked. The banner advertisement corresponding
to the winning bid is displayed to the user. Other advertisements may
be displayed according to their ranking if there is an opportunity to
display additional advertisements (e.g., if the user engages in several
search or browse activities during a session).” )
See e.g., NETGRAVITY ADSERVER CHOSEN BY GNN (“NetGravity, the
leader in Internet advertising technology, today announced GNN, a
242

Reference
GNN

Disclosure
service of America Online Inc., will take advantage of the NetGravity
AdServer technology for WebCrawler. . . . This allows GNN to . . .
dynamically deliver targeted ads. . . . Now, through NetGravity’s
relationship with I/Pro, Web sites will be able to develop and place
advertising much more effectively using management tools with
demographic profiles for targeted ad placement.”)
Lycos, Inc. Registration See LYCOS PROSPECTUS at GOOG-WRD-00872480:
Statement No. 333-354,
dated April 3, 1996
(“LYCOS PROSPECUS”),
produced at GOOGId. at GOOG-WRD-00872492:
WRD-00872476GOOG-WRD00872549

Id. at GOOG-WRD-00872498:

Id. at GOOG-WRD-00872499:

243

Reference

Disclosure

Id. at GOOG-WRD-00872500:

Id. at GOOG-WRD-00872501:

Id. at GOOG-WRD-00872506:

Id. at GOOG-WRD-00872548:

244

Reference

Disclosure

Lycos, Inc. Form S-1
Registration Statement,
dated February 14,
1996 (“LYCOS S-1”),
produced at GOOGWRD-00872550GOOG-WRD00872923

See LYCOS S-1 at GOOG-WRD-00872557:

Id. at GOOG-WRD-00872558:

Id. at GOOG-WRD-00872568:
245

Reference

Disclosure

Id. at GOOG-WRD-00872574:

Id. at GOOG-WRD-00872575:

246

Reference

Disclosure

Id. at GOOG-WRD-00872576:

Id. at GOOG-WRD-00872582:

Id. at GOOG-WRD-00872616:

247

Reference

Disclosure

Excite, Inc. SB-2
Registration Statement
No. 333-2328-LA,
March 11, 1996
(“Excite SB-2”)
produced at GOOGWRD-00872006GOOG-WRD00872094

Id. at GOOG-WRD-00872010.

248

Reference

Disclosure

Id.

Id. at GOOG-WRD-00872011.

249

Reference

Disclosure

Id.

Id. at GOOG-WRD-00872013.

Id.

250

Reference

Disclosure

Id. at GOOG-WRD-00872036.

Id. at GOOG-WRD-00872038.

Id. at GOOG-WRD-00872039.

Id. at GOOG-WRD-00872041.
251

Reference

Disclosure

Id. at GOOG-WRD-00872043.

Id. at GOOG-WRD-00872044.

Id.

252

Reference
Excite, Inc. Prospectus,
dated April 3, 1996
(“Excite Prospectus”)
produced at GOOGWRD-00871928GOOGL-WRD00872005

Disclosure

Id. at GOOG-WRD-00871930.

Id.

Id. at GOOG-WRD-00871931.

253

Reference

Disclosure

Id.

Id. at GOOG-WRD-00871933.

Id.

254

Reference

Disclosure

Id. at GOOG-WRD-00871956.

Id. at GOOG-WRD-00871958.

Id. at GOOG-WRD-00871959.

Id. at GOOG-WRD-00871961.
255

Reference

Disclosure

Id. at GOOG-WRD-00871963.

Id. at GOOG-WRD-00871964.

Id.

256

Reference
InfoSeek Corporation
S-1 Registration
Statement No. 3334142, Amendment No.
1, dated May 3, 1996
(“InfoSeek S-1”)
produced at GOOGWRD-00872371GOOG-WRD00872464

Disclosure

InfoSeek S-1 at GOOG-WRD-00872378.

Id.

Id.

Id. at GOOG-WRD-00872385.

257

Reference

Disclosure

Id. at GOOG-WRD-00872396.

Id. at GOOG-WRD-00872401.

Id. at GOOG-WRD-00872402.

258

Reference

Disclosure

Id. at GOOG-WRD-00872402-403.

Id. at GOOG-WRD-00872403.

Id. at GOOG-WRD-00872404.

Id. at GOOG-WRD-00872404.
259

Reference

Disclosure

Id. at GOOG-WRD-00872404-05.

Id. at GOOG-WRD-00872406.

Id. at GOOG-WRD-00872408.

Id.

Id. at GOOG-WRD-00872410.

260

Reference

Disclosure

Id. at GOOG-WRD-00872411.
Yahoo Prospectus
Registration Statement
No. 333-2142, dated
April 12, 1996 (“Yahoo
Prospectus”) produced
at GOOG-WRD00874251-GOOGWRD-00874328

Id. at GOOG-WRD-00874255.

Id. at GOOG-WRD-00874261.

261

Reference

Disclosure
Id. at GOOG-WRD-00874262.

Id. at GOOG-WRD-00874263.

Id. at GOOG-WRD-00874269.

Id. at GOOG-WRD-00874275.

Id. at GOOG-WRD-00874279.

262

Reference

Disclosure

Id. at GOOG-WRD-00874280.

Id. at GOOG-WRD-00874280.

Id. at GOOG-WRD-00874283.

263

Reference

Disclosure

Id. at GOOG-WRD-00874284.

Id. at GOOG-WRD-00874284-85.

264

Reference

Disclosure
Id. at GOOG-WRD-00874285.

Id. at GOOG-WRD-00874285-86.

Id. at GOOG-WRD-00874286.

Id. at GOOG-WRD-00874287.

265

Reference

Disclosure

Id. at GOOG-WRD-00874289.

Id. at GOOG-WRD-00874292.

Id. at GOOG-WRD-00874292.

266

Reference

Disclosure

Id. at GOOG-WRD-00874327.
Yahoo Form SB-2
Registration Statement
No. 333-2142, dated
March 7, 1996 (“Yahoo
Form SB-2”) produced
at GOOG-WRD00874329-GOOGWRD-00874418

Id. at GOOG-WRD-00874335.

267

Reference

Disclosure

Id. at GOOG-WRD-00874340.

Id. at GOOG-WRD-00874340-41.

Id. at GOOG-WRD-00874341.

268

Reference

Disclosure

Id. at GOOG-WRD-00874343.

Id. at GOOG-WRD-00874348.

Id. at GOOG-WRD-00874357.

269

Reference

Disclosure

Id. at GOOG-WRD-00874358.

Id. at GOOG-WRD-00874358.

Id. at GOOG-WRD-00874361.

270

Reference

Disclosure

Id. at GOOG-WRD-00874362.

Id. at GOOG-WRD-00874362.

271

Reference

Disclosure

Id. at GOOG-WRD-00874362.

Id. at GOOG-WRD-00874363-64.

272

Reference

Disclosure

Id. at GOOG-WRD-00874364.

Id. at GOOG-WRD-00874364.

Id. at GOOG-WRD-00874366-67.

Id. at GOOG-WRD-00874368.

273

Reference

Disclosure

Id. at GOOG-WRD-00874369.

Id. at GOOG-WRD-00874369.

274

Reference

Disclosure

Id. at GOOG-WRD-00874404.
Open Text Form F-1
Registration Statement
No. 33-98858, dated
November 1, 1995
(“Open Text Form F1”) produced at
GOOG-WRD00873727-GOOGWRD-00873878

275

Reference

Disclosure
Id. at GOOG-WRD-00873603.

276

Reference

Disclosure

277

Reference

Disclosure

Id. at GOOG-WRD-00873633-35.

Id. at GOOG-WRD-00873635.

278

Reference

Disclosure

Id. at GOOG-WRD-00873637.

Id. at GOOG-WRD-00873642.

Id. at GOOG-WRD-00873642.

Id. at GOOG-WRD-00873644.

279

Reference
Open Prospectus, dated
January 23, 1996
(“Open Text
Prospectus”) produced
at OT03652-3758

Disclosure

Id. at OT03653.

280

Reference

Disclosure

281

Reference

Disclosure

Id. at OT03689-91.

282

Reference

Disclosure
Id. at OT03692.

Id. at OT03693.

Id. at OT03698.

283

Table B5: Fuzzy Logic
To the extent the references addressed in claim charts A-1 to A-39 does not disclose the
limitations identified in each chart citing Table B5, one of ordinary skill in the art would be
motivated to combine the references addressed in claim charts A-1 to A-39 with any one or more
of the Table B5 references listed below because: it would have yielded predictable results; using
the techniques of the Table B5 references would have improved the primary or obviousness
references in the same way; and applying the techniques of the Table B5 references to improve
primary or obviousness references would have yielded predictable results.
Reference
U.S. Patent No.
6,119,101
(“PECKOVER”)

Disclosure
See, e.g., PECKOVER, 19:3-32:
A Preference Manager function 54 maintains data about the
preferences of the user. Preferences indicate items of interest to
the user, such as favorite brands, interest in sports, etc. Within
Agent System 10, preference data also includes “demographic”
data. Demographic data indicates facts about the user, such as
whether the user is a homeowner, the user’s gender, the user’s
age group, etc. Although marketing industry usage of the term
“demographics” may include a person’s name, address, or
other identifying data, a Preference Manager’s demographic
data does not include data that identifies the particular user.
Preference data may be entered manually by the user using, for
example, a form on a Web page, or data may be loaded by a
System Administrator. Preferences may also be updated
automatically by the system as, for example, when the user
instructs the system to “remember” a product brand name from
a product search. Preference Manager 54 uses preference data
to order search results, so that items that are more likely to be
preferred by the user will be displayed first when the results
are delivered to the user. Referring now to FIG. 5A, each
preference datum 68 comprises not only a value 72, but also a
key 70 for ease of searching. Referring to FIG. 5B, a small
sample of preference data illustrates the kind of data that might
be used. A particular user typically will have much more
preference data. Some values are shown as “rank m in n” to
illustrate that ranking data may also be stored. The specific
keys of any particular set of preference data depends on what
the user has entered, etc. Only keys that are relevant to a
particular user are included in that user’s preferences, and the
284

Reference

Disclosure
specific data maintained will change over time.
PECKOVER, 20:65-21:4:
Referring again to FIG. 4A, a Target Manager function 66
assists the user in identifying Personal Agents to which
targeted ads may be delivered. Target Manager 66 can identify
Personal Agents based on preferences, demographic
characteristics, and Decision Agent activity. Target Manager
66 does not have access to private data of consumer Personal
Agents 12 such as name, address, etc.
PECKOVER, Fig. 5B:

PECKOVER, Fig. 18:

285

Reference

Disclosure

U.S. Patent No.
6,374,237 (“REESE”)

REESE, 3:45-58:
The invention contemplates that the matching server 120
works with the client user profile request 100 to pare down the
data delivered to the client. The matching server 120 preselects an aggregate of data that is determined to be the most
relevant to different sets of user profile requests 100. The
matching server 120 does this by searching various content
sites 130, 140, 150, 160 on the Internet or other network. A
user profile request 100 is applied against the matching server
120 aggregate of data like a sieve, and only data matching the
user profile request 100 is returned to the client 110. The
invention contemplates that the matching server 120 need not
match the user profile 100 exactly, but can accommodate a
user’s designated acceptable range of variability, i.e., a quality
factor.
REESE, 5:55-6:8:
The user profile form 600 includes a Search Type field 630
that allows a user to select whether the user wants an exact
match of the user profile with the search data or whether the
286

Reference

Disclosure
user will accept some lesser amount of exactness as acceptable
for retrieved data. The user profile form 600 further allows the
user to enter demographics specific to the user. In FIG. 6, the
demographics include area code 640, zip code 650, state 660,
sex 670, age 680, and some other identifiers 690. Once the user
enters the appropriate data in the user profile form 600, the
user is instructed to save the profile by a “Save Profile” 694
button. This allows the user to save his user profile and include
the user profile in subsequent searches at subsequent times
without having to repeat the steps of completing a user profile
for each search. Once the form is completed, the user may
submit the user profile by indicating its submission with the
“Submit Profile” 696. In this case, the user profile will be
submitted with the search request as either a POST or GET
method request as specified above with reference to FIGS. 3-5
and the accompanying text.
REESE, 7:53-8:2:
When assessing the database constructed by the matching
server to the user profile, the matching server may require an
exact match or a non-exact match. For an exact match, it is
contemplated that each and every element of the user profile
match that of the data collected in the query database on the
matching server. If such stringent requirements are not
necessary, the user may designate a lesser standard of
stringency and retrieve data that is not an exact match to the
query data and the user profile. In FIG. 9, for example, if the
user profile contained ten distinct data categories, i.e.,
demographic specifics, a user might designate a non-exact
match 934 and then only require a level of stringency 936 of
between 20 and 80 percent matching. If the user demanded
only two of ten elements of the user profile correspond to the
retrieved data, the user might designate a non-exact match of
0.2 or 20 percent. Similarly, if the user wanted 80 percent
accuracy, the user designates 0.8.
REESE, 8:4-24:
Instead of a match/non-exact match system, the invention
contemplates that the retrieved data be associatively matched
to the user profile. For instance, the user profile can specify as
a profile increment “fruit”. The matching server would retrieve
matching data that includes the various kinds of fruits, i.e.,
apples, oranges, etc. In such case, the matching server must be
intelligent to know that an apple or an orange is a “fruit”. It is
known in the art to employ various methods to accomplish
artificial intelligence with computer systems, wherein artificial
intelligence may be described as a system in which a computer
287

Reference

U.S. Patent No.
5,710,884 (“DEDRICK
PATENT”)

Disclosure
is able to reach conclusions based on certain inputs after it has
been trained or instructed in a certain set of rules or
experiences. The most popular artificial intelligence systems
are the so called “heuristic search” models as well as
“associative memory” systems and “connectionist” models. An
associative memory system, for example, solves a current
problem by examining symptoms or characteristics of the
problem and comparing those systems to previous solutions to
the problem. The invention contemplates that an associative
user profile may be implemented with known artificial
intelligent systems.
DEDRICK PATENT, 7:40-52:
When sufficient data has been collected for a particular
consumer variable, then content adapter 25 uses that data to
customize received electronic content to the end user. The
amount of data which is sufficient is dependent on the
particular consumer variable. For example, once personal
profile database 27 has collected ten consumption format
selections from this end user and all ten have been for video
format, content adapter 25 may determine that this is sufficient
data to customize incoming electronic information. However,
content adapter 25 may determine that sufficient data has not
been collected to customize colors if this end user has selected
ten different fields, six of which were purple and four of which
were green.
DEDRICK PATENT, 7:53-64:
In one embodiment of the present invention, the end user is
able to override any compiled user profile data. For example,
even though the end user may select a field with the color
purple most frequently, the end user is able to modify the user
profile data to indicate that green is the preferred color. In one
implementation, the statistic compilation process 26 uses this
input by the end user for its data compilation. Alternatively,
the statistic compilation process 26 may use the data collected
by client activity monitor 24 for its data compilation, or the
statistic compilation process 26 may utilize both the end user
and the data collected by client activity monitor 24.
DEDRICK PATENT, 8:32-15:
In one embodiment of the present invention, statistic
compilation process 26 compiles electronic content-specific
information for return to the metering server 14. This
information includes, for example, how much time the end user
spent consuming the electronic content, and how much of the
content was consumed. For example, a particular
advertisement may include ten different screens which are
288

Reference

Disclosure
displayed to the end user. If the end user spends 15 seconds
viewing the first screen and 15 seconds viewing the second
screen and then tenninates the advertisement, the statistic
compilation process 26 transfers information to the metering
server 14 indicating that an individual with this end user’s user
profile data spent 30 seconds viewing the electronic
information and that the content was 20 percent consumed
(that is, two screens out of ten were consumed). Additionally,
information indicating the specific elements of the
advertisement that were consumed (that is, the first two screens
in this example) is also transferred to the advertiser. Note that,
as discussed above, this aggregate information does not reveal
the identity of the end user who consumed the advertisement.
DEDRICK PATENT, 9:28-45:
When requesting electronic advertisements, the data returned
to the end user is dependent on the end user’s request. For
example, the end user may define certain results which should
occur based on how well the electronic information matches
the search criteria. The appraisal agent 28 may electronic ed to
return the title of the electronic advertisement if it is only a 5%
match to the search criteria, an abstract if it is a 25% match to
the search criteria, and the entire advertisement if it is a 95%
match to the search criteria. Alternatively, the appraisal agent
28 may be programmed to return only titles, regardless of how
well the advertisements match. In addition, the appraisal agent
28 may know, based on the user profile data stored in personal
profile database 27, that the end user only wants to consume
five electronic advertisements per day. The appraisal agent
may then retum titles of 25 electronic advertisements to the
end user, and allow the end user to select which advertisements
he or she will consume.
Wilms, A Natural
WILMS, p. 3:
Language Interface For
The natural language interface catches obvious misspellings
An Intelligent
and employs fuzzy logic techniques to automatically translate
Document Information
user specifications like “very”, “especially,” or “not” into
And Retrieval System
weights. The interface also employs a transparent synonym
(1988) (“WILMS”)
lookup to improve category matching.
WILMS, p. 12:
However, an interface based on key word matching and fuzzy
set techniques is proposed, which is able to handle relatively
unconstrained natural language queries and thus eliminate the
need for mastering a formal query syntax.
WILMS, p. 17-18:
The chronology base also contains synonyms (“after” =
“beyond” = “past” = “since”), and establishes concrete
289

Reference

U.S. Patent No.
7,072,849 (“FILEPP”)

Disclosure
values for fuzzy specifications (“recent” = after 1986) (see
Figure 4). Many of these concrete values are dynamic, and
depend onthe current year (recent means different things in
1987 than in 1989) and on the oldest document in the
collection (if the oldest document was published in 1957 or in
1976 “earliest papers” takes on quite a different meaning). It
may even mean different things to different users (i.e., while
“recent” means “the last two years” for one researcher, it may
mean “the last two months” for another. The value of “now”
(as in “all papers from 84 till now”) also depends on the
current year, of course. It may even be possible to retrieve
“new” documents, if the system keeps track of updates to the
document collection since the last interaction with the IIRS.
When intensifiers are used in combination with fuzzy
specifications (e.g., “very recent”), the interface uses a
dynamic weighting scheme (e.g., 1986 (0.6) 1987 (0.8) 1988
(1.0)) (See Chapter Four).
WILMS, p. 37-38:
These search terms consist of “crisp” items (“marketing,”
“practice”), imprecise terms (“recent”), and fuzzy quantifiers
(“very”). The last two are considered fuzzy because they
convey imprecise information, and do not have sharp
distinctions between membership or non-membership. To
handle these uncertainties, each concept is given a weight,
which is determined by fuzzy logic [ZADEH 81]. These
weights range between -1.0 and1.0, and are used by the
retrieval component in addition to weights stored in the
inverted files to identify relevant documents (see step 6 in
Figure 8).
See, e.g., FILEPP, 21:19-34:
If the string entered by the user matches a keyword existing on
one of the keyword tables, and is thus associated with a
specific PTO, RS 400 fetches and displays associated objects
of the partitioned applications and builds the entry page in
accordance with the page composition dictated by the target
PTO.
If the string entered by the user does not match a specific
keyword, RS 400 presents the user with the option of
displaying the table of keywords approximating the specific
keyword. The approximate keywords are presented as
initialized, cursorable selector fields of the type provided in
connection with a Index command. The user may then move
the cursor to the nearest approximation of the mnemonic he
originally selected, and trigger navigation to the PTO
associated with that keyword, navigation being as described
290

Reference

Another Search
Engine? Hotwired
Introduces Hotbot,
Powered By Inktomi,
PR Newswire, May 20,
1996 (“ANOTHER
SEARCH ENGINE”)

https://web.archive.org/

Disclosure
hereafter in connection with the RS 400 native code.
FILEPP, 34:25-39:
Data collection manager 441 gathers information concerning a
user’s individual system usage characteristics. The types of
informational services accessed, transactions processed, time
information between various events, and the like are collected
by data collection manager 441, which compiles the
information into message packets (not shown). The message
packets are sent to network 10 via object/communication
manager interface 443 and link communications manager 444.
Message packets are then stored by high function host 110 and
sent to an offline processing facility for processing. The
characteristics of users are ultimately used as a means to select
or target various display objects, such as advertising objects, to
be sent to particular users based on consumer marketing
strategies, or the like, and for system optimization.
See, e.g., ANOTHER SEARCH ENGINE, p. 1: “’The rules of the search
engine game have changed. Internet users thought they’d get what
they needed from traditional search engines, but they found the result
to be thin on content, rigid in context, and often totally irrelevant,’ said
Andrew Anker, president and CEO of HotWired Ventures. ‘Our quest
to find a better search engine led us to Inktomi. By combining the best
technology, the most relevant searches, and an innovative interface,
we created HotBot -- a bigger, better, smarter way to search the
Web.’”
ANOTHER SEARCH ENGINE, p. 2: “HotBot includes a number of unique
features. Users can get the most current information quickly,
efficiently view and use that information, and interact with the search
engine in a personal manner. Daily Updates: The HotBot spider crawls
the Web every day, offering users the most current information.
Reliable and Fast: HotBot's fault-tolerant engine reliably delivers
query results in seconds, without frequent downtime. Convenient
Previews: HotBot allows users to preview documents without leaving
the search page, reducing search time. Personal Searching: The
HotBot interface allows users to personalize their search engine to fit
their own surfing style.”
ANOTHER SEARCH ENGINE, p. 2: “HotBot identifies, customizes, and
ranks millions of Web documents using an algorithm developed by a
team of the world's leading experts in information retrieval. HotBot
recognizes that users desire varying levels of information detail, so it
allows users to control the amount and type of information searched.
The computing power available to HotBot enables the user to define a
search query using a wide range of criteria in a way that is not possible
with more traditional search engines.”
The first commercial application of Inktomi's innovative technology is
291

Reference
web/1996110
6235936/http://
www.inktomi.com/

Sadaaki Miyamoto,
“On Fuzzy Information
Retrieval,” Japanese
Journal of Fuzzy
Theory and Systems,
Vol. 3, No. 1 (1991)
(“MIYAMOTO”)

Development of the
Coder System: A
Testbed for Artificial
Intelligence Methods in
Information Retrieval
(“Fox”)

Disclosure
the HotBot™ search engine service, offered in conjunction with
HotWired, Wired magazine's electronic sibling. By leveraging this
scalable technology, HotBot was the first search engine to index and
search the entire World Wide Web, and represents the only search
engine technology in existence that can expand to match the Web's
growth as it doubles and doubles again.
SmartRelevance. Based on algorithms developed by informationretrieval experts at the University of California at Berkeley, HotBot's
SmartRelevance technology exploits syntactic clues in documents and
relationships between documents, to rapidly identify the most
meaningful information.
MIYAMOTO, e.g., p. 93, “The book by Salton and McGill (1983) is a
basic introduction to the field which divides the study of databases and
information retrieval into five areas: (1) information retrieval, (2)
database management systems, (3) operational information systems,
(4) decision-making assistance, and (5) query response systems.
Information retrieval also includes the study of scientific documents.
We will pay attention to the above classifications while discussing
fuzzy information retrieval. . . The study of fuzzy information retrieval
was begun in the early 1970s. It was not until the 1980s that
realization of fuzzy information retrieval seemed promising. To
accomplish this it was necessary to have faster hardware, software,
and database storage, and propagation of workstations, databases, etc.
The importance of fuzzy information retrieval is now understood by
researchers, who are primarily concerned with ordinary documentary
information retrieval. This was made possible by the clearly stated
methodical framework of fuzzy theory (Zadeh, 1973).”
See, e.g., FOX, p. 349:
Fifth, it is possible to combine natural language processing (as
in group 4) with special query evaluation methods. CALIN,
IOTA, and PROBIB-2, all mentioned above, have a natural
language query-handling capability and distinctive document
representation schemes. In addition, Biswas et al. [1 19,120],
in their work on knowledge-assisted document retrieval,
consider both the natural language interface and the retrieval
components. They developed a modular design, and plan to
carry out a variety of experiments with the System. Their
natural language interface can handle a restricted query
sublanguage through its augmented transition network and can
determine the number of documents desired, the time range of
interest, and the subject matter or content [I19]. The retrieval
component uses fuzzy set theory and one of several
combination of evidence schemes [I20].
FOX, p. 351:
292

Reference

Disclosure

FOX, p. 352:
Retrieval is prompted by an explicit (or default, from the user
model base) query. User model building, problem state
transformation, and building of the problem description all
proceed. When some terms are available, the lexicon can be
accessed by a term expander to obtain other related terms that
can be browsed or automatically used to help construct a
query. Eventually a p-norm or other query is constructed, a
search is made, and a report is prepared for the user.
FOX, p. 352:
Since development of CODER involves research assistants,
students working on MS projects, and students completing
class proiects, it is difficult to characterize precisely the status
of implementation. The initial lexicon, the knowledge
administration complex, the blackboard/strategist complex, the
communications enhancements to MUProlog, a time/date
handler, a p-norm search expert, and two versions of the user
interface manager do function and are all nearly complete.
initial versions of the document-type expert and the user model
builder are being further developed. The document analyzer
and some of the specialists it uses are partially complete.
FOX, p. 357:
5.2. p-norm search expert
The p-norm query notation, which extends Boolean
expressions to allow relative weights to be attached to terms
293

Reference

Architecture for AgentMediated Personal
News Service
(“TURPEINEN”)

Disclosure
and clauses and which allows “p-values” on the AND and OR
operators to indicate the strictness of interpretation of the
operation, is discussed in a work by Salton et al. [I42]. While
other schemes for “soft Boolean evaluation” have been
proposed [I43], none has been shown to perform as effectively
as the p-norm method [i44]. p-norm query processing has been
incorporated in both the SMART and SIRE systems [I45].
Because of its expressive power, the p-norm query form has
been adopted in CODER as one of the canonical query forms.
As can be seen in Fig. 8, a p-norm search expert has been
developed that supports calls through the blackboard to attend
to the “pnorm_query” area. The result of normal processing is
to generate hypotheses for documents best satisfying the query
expression, estimating the degree of relevance they have to the
query.
TURPEINEN, p. 3:
Agents can be considered as mediators [Wiederhold92] that
refine and forward information from heteregenous data sources
to the users. Multi-agent intercommunication methods enable
message passing between agents in a network environment.
The consumer agent transmits user requests for potential
producer agents and filters messages according to user
preferences. The producer agent acts as an information broker
that has a domain model of its own expertise [Fikes95]. The
producer agent can advertise the services to the consumer
agents in the network. Agents negotiate how, when, and which
information items should be transmitted. Agents are also able
to consult other agents for suggestions and further information.
Finally the agents assist in completing necessary data transfer
tasks and financial transactions.
TURPEINEN, p. 6:
1. User modeling. Consumer's preferences are maintained in a
user model. The maintenance can be done explicitly by the
user or automatically a by a learning mechanism in the
consumer agent.
2. Content queries and promotion. Consumer agent sends a
query to the producer agent to receive items that match the
user interests. Also parts of the consumer's user model can be
sent to be used in social information filtering performed by the
content producer. Producer advertises its services to consumer
agents.
TURPEINEN, p. 9:
The system uses a combination of content-based filtering and
social filtering techniques [Malone87, Shardanand95]. The
news selection service is based on a user profile that
294

Reference

Disclosure
consists of:
• keyword-based query profile on user-specified topics;
• semantical matches based on predefined categories;
• trusted agents that send recommendations to each other.
TURPEINEN, p. 10:
The keyword-based selections are defined entirely by the user.
These are normally used to cover short-term information
needs. Each topic is identified by a topic header, producer
agent and a collection of keyword/weight -pairs. The keyword
weight is measured as a value in the range between 0 and 1.
The weight can be adjusted by the user or by the learning
module of the consumer agent. Also exclusive keywords can
be entered to discard articles.
TURPEINEN, p. 352:

P. Bosc, “Fuzzy
querying in
conventional
databases,” Fuzzy Logic
Management of
Uncertainty (1992)
(“BOSC”)

BOSC, e.g., at 646-47, “We now make prech;e the meaning of
"flexibility" assumed in the following. A system is flexible in so far as
it allows imprecise terms in user queries. Consequently, it becomes
necessary to determine to what extent a certain element matches more
or less the query more than another element, which leads to a
classification or ranking of the selected elements. According to this
dermition, we are essentially concerned with items 4 and 5 of the
above list. However, since very often an implicit objective is to avoid
empty answers, the approaches reported hereafter are also connected
with cooperative answers. Several approaches allowing imprecision in
user queries can be imagined and some of them have been proposed
and implemented in research prototypes. One idea is to consider
queries made of two parts: a Boolean qualification selecting elements
295

Reference

Disclosure
and an imprecise condition intended for the ranking of these elements.
Another approach is to allow imprecise queries. Then, two main cases
appear depending on the interpretation of imprecise conditions. As a
matter of fact, we can imagine translating an imprecise condition into
a Boolean one expressing intervals of acceptance and such that some
kind of "distance" is computed for each selected element. An alternate
view is to use fuzzy sets as a basis for the evalution of imprecise
conditions. Here again, some kind of distance is computed for each
element, but this framework is more general than the previous one. In
fact, we shall see that the central point of a system depends on
whether or not it is based on the Boolean logic.”

Mark Lager, “Spinning
a Web Search,” (1996)
(“LAGER”)

LAGER, e.g., “The presentation is targeted toward WEB searchers, in
particular, reference librarians and those who navigate the Internet on
a frequent basis. This presentation will look at search engines,
comparing search techniques and noting differences. The workshop
will identify
use of new computing strategies for information retrieval within each
engine.”
LAGER, e.g., “As Brian Pinkerton states, "The World Wide Web is
decentralized, dynamic and diverse; nativagion is difficult and finding
information can be a challenge." (Pinkerton, 1994). The useful and the
innocuous are lumped together in this huge collection. Academic
information
(e.g., journal articles and course materials) is combined with social
culture information and with
personal home pages. There is no separation. Mark Nelson calls this
information anxiety - the
overwhelming feeling one gets from having too much information or
being unable to find or interpret data. (Nelson, 1994). To be of any
information value, the data must first be organized and retrievable,
providing some structure. Search tools have begun to put some
organization to these uncharted waters. Current trends in information
retrieval offer better opportunities to make more efficient use of this
information resource.”
LAGER, e.g., “The search engine provides more control for the user in
performing a search. Engines use the index to fetch terms of the query.
This means that the more data in the index, the higher the recall.
Indexing every word or the most used words can lead to higher recall
depending on the search query. The larger the index, the more
possibility of hitting upon the words of the query. And, with the size
of the Web, the more often the index is updated, the greater the
number of hits. Search engines on the Web incorporate a number of
296

Reference

Disclosure
techniques to assist in both recall and precision. There are search
engines that employ traditional methods like thesauri or Boolean
searching. Rather than being only a keyword search, the engine will
make logical connections to a thesaurus to enhance recall. Using
Boolean logic (and, or, not, adjacency operators) search engines can
assist in making the query more precise. Different engines have
different defaults.
Natural Language Processing: Relevancy feedback/weighing
probabilistic logic: query by example
fuzzy logic: query expansion
Bayesian networks: case-based reasoning
parallel computing (Inktomi): concept based searching”
LAGER, e.g., “Will it rain today? What is the possibility of my car
needing an oil change? Or, what is the chance of getting an A on my
history test?. There are many questions like these that cannot be
answered with an affirmative or negative answer. Uncertainty reigns.
In an effort to make a decision which accounted for such doubt, in the
midst of chaos, a branch of logic was defined to study probability.
Since the 16th and 17th centuries, probability theory has been used
to explain chance. Such questions rely on a factual information as
history coupled with probability. In information retrieval, the same
applies. By setting up a formula, an algorithm, that places values on
words, their interrelationships, proximity, and their frequency, the
computer can be used to help locate relevant sites. By computing these
terms together, the search engine can produce a relevancy ranking that
is then displayed to the user. (De Bra, 1995) Probabilistic logic is
founded on the presumption that certain factors can be established
logically and mathematically to focus a search. It is similar to fuzzy
logic where the central notion is that truth values (in fuzzy logic) or
membership values (in fuzzy sets) are indicated by a value on the
range [0.0, 1.0], with 0.0 representing absolute Falseness and 1.0
representing absolute Truth. (Brule, 1985)”
LAGER, e.g., “A survey of the Search Engines available from
Netscape's Net Search will help in explaining some of the techniques
discussed. By conducting a search for current trends in information
retrieval, differences can be seen in the structure and techniques of
each engine.
Alta Vista {http://www.altavista.com/}
Techniques and features
Boolean - must use and, or, not, near (10 words) in Advanced Search
Allows user-influenced results ranking
Ranking: title words or first few words
297

Reference

Disclosure
Closer to each other
Document has more of the words
More copies of the words throughout
Parentheses for nesting
Can restrict to field (qualifiers)
Excite http://www.excite.com/
Techniques and features
Concept based searching-use statistical strength of interrelationships
between words
Creates its own knowledge base (or internal thesaurus)
QBE - "similar documents"
Boolean searches
Keyword searches
Relevance - marked with red X
Robot is called Architext
Infoseek {http://infoseek.go.com/}
Techniques and features
Weight terms (required, desirable, undesirable)
Similar pages - QBE
Boolean operators
Natural language
Search mechanisms
Lycos {http://www.lycos.com/}
Techniques and features
Probabilistic retrieval
Indexes top 100 words and 20 lines of abstracts
Keyword searching
Boolean searching
Automatic truncation
Spinning a Web Search : Trends in Information Retrieval Page 7 of 10
http://misc.library.ucsb.edu/untangle/lager.html 4/22/2014
Adjacency 0.0 - 1.0
Results categorized
Terms in bold
Relevancy: early on vs. farther down
Magellan {http://magellan.excite.com/}
Techniques and features
Reviewed by writers
Boolean searching
Green light for information for all age groups
Web, ftp, gopher, newsgroups, telnet sites
298

Reference

Disclosure
Browse directory or Use search engine
Relevancy = frequency of words
Browse button
Robot named Verity
Lists up to 20 pages at the bottom of the screen
Open Text {http://www.opentext.com/omw/f-omw.html}
Techniques and features
Boolean searching
Field operators: anywhere, summary, title, first heading, URL
Query-by-example.”

LAGER, e.g., “Information search and retrieval is of major importance
in locating relevant materials. The ability to aid and assist a user in
finding relevant information is the goal of librarians and information
scientists. On the Web, search engines have made the pr ocess easier
by incorporating a number of newer techniques which include
artificial intelligence, Bayesian statistics and probability theory,
weighting, and query by example. With the goal of finding relevant
materials, these new techniques locate infor mation and also refine the
search query. Since search engines have different criteria in creating
the indexes, it is most useful to use more than one engine in searching
the Web to gain relevant information. As a rule, the more critical or
focused the q uery, the more engines should be applied. With
advances in the tools for information retrieval, the future holds
exciting possibilities for searching on the World Wide Web.”
Henrik Larsen and
LARSEN II, e.g., Abstract, “The problem solving strategy applied in
Ronald Yager, “The
knowledge based systems may ofted be characterized as classification.
Use of Fuzzy Relational Central to classification is computation of the degree to which an
Thesauri for
object is an instance of a given class (concept, category). Two kinds of
Classificatory Problem problems are distinguished, object-querying and classquerying, as
Solving in Information exemplified by, respectively, information retrieval systems and expert
Retrieval and Expert
systems. In the first kind, the problem is to identify the objects (e.g.,
Systems,” IEEE
documents) to which a given concept (the query) applies. In the
Transactions on
second kind, the problem is to identify the concepts (categories) that
Systems, Man, and
apply to a given object (the observation). A fuzzy-set-based scheme
Cybernetics, Vol. 23,
for construction of efficient problem solving systems of the two kinds
No. 1 (Jan./Feb. 1993)
is developed. The problem of vocabulary mismatch is considered in
(“LARSEN II”)
information retrieval, and introduce the scheme as a solution to this
problem. The knowledge base applies a term-centered representation
form called a “fuzzy relational thesaurus.” To avoid recomputation of
deductive information in problem solving tasks, we derive initially the
deductive closure of the knowledge base. This closure is computed in
O(n3) time as the transitive max-star closure of the fuzzy implication
relation represented by the knowledge base; n is the number of terms
299

Reference

Disclosure
in the knowledge base. An upper bound for the closure is computed in
only O( m log m ) time by an algorithm that pai-titioning the terms
into similarity classes; m is the number of pairs of terms for which a
relationship is represented in the knowledge base.”

Peretz Shoval, “ERSE :
An Expert Retrieval
System for Electronics
Databases,” in Expert
Systems for Information
Management, Vol. 3,
No. 2 (1990)
(“SHOVAL”)

SHOVAL, e.g., Abstract, “This paper describes an expert system for
information retrieval in electronic databases: ERSE. The objective of
the system is to support engineering professionals in formulating
proper queries and submitting them to a retrieval database. The system
consists of: (a)a knowledge-base, which is a thesaurus of terms and
semantic relationships, implemented as a semantic network; (b) a
search and evaluation mechanism: the inference-engine, which
conducts a guided search aimed at finding appropriate query terms.
While doing so it invokes relevant knowledge, evaluates it, and
suggests final findings to the user; (c) a database of patents in the
domain of error-correction codes, implemented with a relational
database management system (DBMS); (d) a retrieval mechanism,
which measures the similarity between the system generated weighted
query, and the index terms of patents, and returns a rank-ordered set of
patents. The user is then able to provide feed-back and improve his
query accordingly; (e) user interfaces, including system capability to
explain its findings/decisions. The system is implemented in Pro log,
C and INGRES, under Unix . The system design is described, and
examples of its operation and evaluation of its performance are
given.”
SHOVAL, e.g., at 88-90, “ERSE takes this fourth approach for
integrating expert and IR systems. However, ERSE is actually a
complete system for information retrieval, including the other
components: an interface, a database, and a retrieval mechanism.
Other major features of the system are:
(a) It accepts a user query composed of a list of weighted terms, and it
generates an appropriate weighted query which is submitted to a
retrieval database. Consequently it returns a set of rank ordered
documents, using a fuzzy approach.
(b) The thesaurus is built as a semantic network which is particularly
suitable for representing declarative, conceptual knowledge.
(c) Both the database of documents, and the knowledge-base of the
expert system, are stored in a relational database. This allows the
handling of large databases and knowledge-bases.
(d) A convenient user interface is provided, enabling the system to
explain and justify its decisions, and enabling the user to feedback and
affect system behaviour.
ERSE is based on and extends an earlier system that has been
300

Reference

Disclosure
developed by the first author for the domain of business
administration11,12 and that has since been utilised in medicine. The
principles and components of that system are described in Section 2.
The domain used for our system is electronics, and specifically patents
in error-correction codes technology. This domain is typical of many
technological areas in which users often have specific and directed
information needs. Section 3 discusses the specific information
retrieval needs in electronics, which led us to develop ERSE. Section
4 details the architecture, components and processes utilised in ERSE,
emphasising the specific contributions it makes, compared to its
predecessor. Section 5 presents an annotated example of the system
operation. Then Section 6 presents an evaluation of the performance
of the system, based on a set of case-study queries. Section 7
highlights some implementation issues, and Section 8 points to future
developments.

Sameer Singh, “Fuzzy
Pattern Recognition for
Knowledge-Based
Systems,” Proc. 6th
International
Conference on Data
and Knowledge Systems
for Manufacturing and
Engineering
(DKSME'96), Tempe,
Arizona, USA, pp. 110, (24-25 October,
1996) (“SINGH”)

SINGH, e.g., Abstract, “Knowledge-based systems have been severely
restricted in areas where the speed of processing is a key factor. This
is especially evident in large systems where the speed of knowledgebase searches is important. This paper proposes a fuzzy pattern
recognition technique which identifies data patterns using possibility
distributions and documents a fuzzy algorithm which is implemented.
The technique is based on the theory of possibility. The results
obtained using sensor data in manufacturing are encouraging: the
fuzzy technique outperforms non-fuzzy techniques convincingly. The
results for comparison with non-fuzzy techniques include shell-sort
and quick-sort with binary search. The fuzzy technique identifies the
correct pattern in the sensor database with nearly 99% accuracy. These
results highlight the role of new fuzzy technologies for making
knowledge-based systems more attractive in areas where they are
currently limited by speed considerations.”

Lotfi Zadeh, “The Role
of Fuzzy Logic and Soft
Computing in the
Conception and Design
of Intelligent Systems”
(“ZADEH”)

ZADEH, e.g., Abstract, “As one of the principal constituents of soft
computing, fuzzy logic is playing a key role in the conception and
design of what might be called high MIQ (Machine Intelligence
Quotient) systems. There are two concepts within FL which play a
central role in its applications. The first is that of a linguistic variable,
that is, a variable whose values are words or sentences in a natural or
synthetic language. The other is that of a fuzzy if-then rule in which
the antecedent and consequent are propositions containing linguistic
variables. The essential function served by linguistic variables is that
of granulation of variables and their dependencies. In effect, the use of
linguistic variables and fuzzy if-then rules results – through
granulation - in soft data compression which exploits the tolerance for
imprecision and uncertainty. In this respect, fuzzy logic mimics the
301

Reference

Disclosure
crucial ability of the human mind to summarize data and focus on
decision-relevant information.

Donald Kraft,
“Research into Fuzzy
Extensions Retrieval”
(“KRAFT”)

KRAFT, e.g., Abstract, “Modern computerized information retrieval
systems consist of mechanisms to acquire, describe (e.g., index), and
store "documents", and to receive, analyze, and respond to queries for
information for users. A key element is the index language, by which
the users (or user intermediaries) and indexers can communicate.
Modern technology allows natural language processing mechanisms to
begin to be incorporated in the sense of matching terms found in the
free text specification of the query and the free text within the
document.
Various models of retrieval have evolved over time. The vector space
model treats both documents and queries as points in the Cartesian
space formed as the product of all possible index terms. Then,
documents deemed "near" the query, i.e., "similar" to the query, are
retrieved. Work has been done on clustering "similar" documents to
facilitate the retrieval processing. A second model incorporates
probability into the retrieval system by attempting to
evaluate the likelihood that each document is relevant to a given
query.”
KRAFT, e.g., Abstract, “One key element in both these approaches is
that the index terms assigned to the documents can be weighted. These
weights may be derived from relative frequencies of term occurences
or from subjective estimates of likelihood of relevance. Another
weighting scheme can be considered that allows the incorporation of
Boolean logic into the query mechanism; that of fuzzy set theory. This
theory lets the concept of imprecision be entered into the model, and is
well-known, albeit controversial.”
KRAFT, e.g., Abstract, “One can extend the fuzzy Boolean model by
generalizing to weights being assigned to the query terms as well. This
can cause problems with the fuzzy Boolean l a t t i c e , however. One
must consider such c r i t e r i a as s e p a r a b i l i t y , generalization,
and self-consistency when designing query processing mechanisms. A
mechanism for considering these query weights as thresholds solves
some of these problems, but the semantics of the weights, especially
as the low end, is not clear.”
KRAFT, e.g., Abstract, “Extensions of the vector space and probability
models have been considered by other researchers to try to incorporate
Boolean logic. In addition, this has allowed consideration of adding
relevance feedback to the fuzzy Boolean model. Another issue is to
generate means of evaluating a fuzzy Boolean retrieval system. One
302

Reference

Disclosure
can obviously try to generalize recall and precision. However, the real
problem is to properly incorporate rank ordering, which the weighting
and fuzzy query processing mechanism provide.”

G. Bordogna et al.,
“Fuzzy Inclusion in
Database and
Information Retrieval
Query Interpretation”
(1996) (“BORDOGNA”)

BORDOGNA, e.g., Abstract, “Abstract. In this paper, a short review of
the role of the inclusion operator in the interpretation of queries
addressed to databases and Information Retrieval Systems (IRSs) is
analyzed. Some properties and semantic aspects of various
definitions of fuzzy, inclusion are discussed and applied to interpret
queries in Data Base Management Systems and IRSs”
BORDOGNA, e.g., at 548, “This basic model of IR has been extended to
tile main aim of
providing a flexible matching mechanism able to evaluate the degree
of relevance or satisfaction of each retrieved document with respect to
the query. These models are based on two main ideas the association
of a weight with both each term in the representation of documents
(index term weights) and each term in the query (query term weight).
Index term weights express the significance of terms in representing
the document contents, while query term weights indicate the
importance that terms should have in the desired documents.”
BORDOGNA, e.g., at 551, “In this paper the role played by the inclusion
operation in both
the division of fuzzy relations in DBMSs mid in weighted query
evaluations in extended Boolean Information Retrieval is investigated.
Some fuzzy approaches presented in the literature
are reformulated in the unified framework of fuzzy inclusion. Future
developments of this work will cone, era two points: i) the weakening
of the universal quantifier implied in the division of fuzzy relations
and ii) the consideration of more general queries (not only
conjunctive) in IRSs.”

“Automatic Thesaurus
Construction
Supporting Fuzzy
Retrieval of Reusable
Components,” (1995)
(“DAMIANI”)

DAMIANI, e.g., Abstract, “Effective access to repositories of reusable:
components should rely on retrieval functionalities based also on
imprecise queries. This paper presents a fuzzy retrieval model based
on keywords describing the functionalities of reusable components.
Fuzzy weights are assigned to these keywords automatically. Retrieval
is supported by a Thesaurus where a fuzzy synonymia relationship is
used to c:ompute adaptability of reusable components to the needs
expressed by the user fuzzy query. The adaptability index is
ameliorated along time via a quality function reporting feedback on
the system usage.”

303

Reference

Disclosure
DAMIANI, e.g., at 542-43, “Software reuse needs effective retrieval
techniques to make development with reusable components more
convenient than development from scratch [Kru92]. To th,is aim,
components should be appropriately described and user ‘queries
should allow for a degree of uncertainty in order to isolate a set of
components that can be adapted to the new application [Pri93]. Many
of the existing software libraries or repositories exhibit both the
classification problem (description of components), and the retrieval
problem [Ban93. Bat92. Dev91].
This paper proposes a technique for Thesaurus-based software
retrieval from a repository. The technique D based on software
descriptors containing keywords weighted with fuzzy values to
describe the behavioral proprties of reusable components. Central to
this approach is Thesaurus automatic construction starting from the
software descriptors. The approach supports imprecise queries through
the use of fuzzy logic [Kli88, Kos92].
The descriptors are assumed to be constructed from. the code and from
its accompanying documentation. The object oriented SIB (Software
Information Base) repository is considered, whose descriptors are
classes [Con93]. SIB classes have a usual class attribute part, and an
additional keyword-list based part at the basis of the retrieval model.
The model is given in terms of weighted pairs of related keywords
(features) interpreted as open class keywords [Maa describing the
component functionalities. The fuzzy weight associated to each
feature expresses the degree of imprecision that characterizes the
description. For the fuzzy weights in the SIB, the paper proposes an
assignment algorithm employing a Feature Weighting Function (FWF)
adapted from a classical term weighting function used for document
retrieval [Sal88].
The retrieval model enables to pose imprecise queries, asking for a set
of characteristics expected from the component. Imprecise queries are
lists of features, describing the characteristics of the needed
component and fuzzy weight in the query specifying how relevant
each feature is for the developer. Returned candidates are ranked
according to their degree of aduptabilify to the required
limctionalities. Retrieval is assisted by a Thesaurus containing unique
terms and synonyms. Terms are single keywords taken from
descriptors in the SIB. or added by the Application Engineers in
charge of SIB maintenance. Terms are organized in the SIB and in the
Thesaurus by conrexrs (or facets [Pri87] or categories [Gib90]) acting
as search environments. A term in a context gets a fuzzy value of
relevance representing how significant the term is in that context. The
automatic construction of the Thesaurus consists in extracting terms
from the SIB descriptors and in computing the fuzzy relevance of each
term in the contexts using a Context Relevance Function (CRF).
304

Reference

Disclosure
Terms are linked to one another via a fuzzy synonymia relationship
which is interpreted as the adaptability index of the software
components described by the synonyms. The adaptability is computed
dynamically taking into account the fuzzy weight of synonymia
between terms in the ThesaurusBoth the SIB weights and the
Thesaurus weights are mantained automatically and ameliorated along
the system life cycle employing a Quality Function (QF) which
observes the user reactions to query answers from the system, and in
batch mode, slowly variates the fuzzy weights.
A QF is proposed in the paper which. exploiting the theory of fuzzy
sets, implements an adaptative retrieval system (Kli88. Mun94],
tunable with use along time.
The paper is organized as follows: a general view of SIB descriptor is
provided with the technique of fuzzy weighting of descriptors. The
concept of adaptability between components is defined and applied to
retrieval. Automatic Thesaurus construction and syonymia
computation are presented.”
DAMIANI, e.g., at 546, “This paper has presentes an approach to fuzzy
retrieval of
components from a repository based on fuzzy-weighted keyword pairs
(features) describing the component functionalities. A method for
automatic assignment of weigths to features has been described. The
approach relies on a Thesaurus of terms used to describe the reusable
components. The basic relation in the Thesaurus is synonymia which
is also fuzzy; its connection to the retrieval model has been shown and
a method for automatic construction of fuzzy synonyms in the
Thesaurus has been illustrated. At the user interface level a prototype
[Fau93] has been experimented using ranged values to simplify the
user interaction. Evaluation of the software retrieval operations is
undergoing using code and design documents. In particular. assuming
that “nice” features are contained in the SIB descriptors and suitable
contexts are initialized within the Thesaurus, we are evaluating the
approach using a library of object-oriented code and a library of
conceptual application schemas based on the E/R model.”

Duncan Buell,
“Performance
Measurement in a
Fuzzy Retrieval
Environment,” 1981
(“BUELL”)

BUELL, e.g., at 56, “Measuring the extent to which a computerized
document retrieval system fulfills the goals set for the system is a
complex problem that involves everything from initial goal
specification to the actual underlying computer software. An average
user will view the system as a "black box." The user makes requests;
the system responds. Numerous factors will thus affect the evaluation
of the system by such a user. These include such varied aspects as
physical ease of use, the user's ability to understand how to formulate
requests, the coverage of the desired topic by the collection (and the
305

Reference

Disclosure
coverage of new or old material at the user's appropriate level), and
even the user's own knowledge of the topic in which he is interested.
We emphasize that we are investigating only a narrow aspect of
retrieval system evaluation. We consider not the "human engineering"
required to provide the average user with the information he desires,
but the establishment of quantitative standards by which to measure
the ability of the mathematics and logic of the retrieval decision
mechanism to select for retrieval in response to a request the same set
of documents which would have been selected by a human expert
unaided by the automated system. [9, 16, 18] Among these standards
are recall and precision and associated measures and various measures
of "value" returned in comparison to the search length. These
measurements are well-defined for systems with Boolean indexing
and standard Boolean query-to-document matching functions.”
BUELL, e.g., at 58, “One problem which immediately arises in
measuring performance is that, if the RSV's are no longer simply 0
and i, then a new interpretation must be made of "about" and of
"retrieved." The first problem is resolved if a set-theoretic, indeed
fuzzy settheoretic, interpretation can be placed on all numerical values
involved. [7, 20] In a system in which RSVIs are not simply 0 and i,
however, it is no longer the ease that one would simply retrieve a
subset of the documents. The user might instead be given information
on a ranked list of documents, for example, and asked to specify a
threshold above which to actually retrieve. Or, following the ideas of
Cooper [6], the user might be given the ranked list and allowed to
retrieve one document at a time until he decided that he had seen
enough . There are several possibilities; the problem remains the
same--the set RT is definable not by the RSV, but by system
convention or,
worse yet (from the point of view of predictability for use in
numerical measurement),
by user whim. By a generalized retrieval system, then, we shall mean
a system in which either
1. the indexing function is Boolean, the queries resemble Boolean
queries, but the RSV's are not Boolean (this would include retrieval
mechanisms such as the cosine coefficient); or
2. the indexing function is fuzzy, queries resemble Boolean queries,
and RSV computations follow normal fuzzy subset rules (this would
be a simply fuzzysubset system, as described by Sachs, Tahani, and
others [i0, ii, 15, 17]); or
3. the indexing function is fuzzy, the queries have weights or
thresholds attached to terms and/or subexpressions, and RSV
computation is not necessarily simply a fuzzy-subset MF computation
(this would include systems such as those suggested by Bookstein [i],
306

Reference

Disclosure
Buell and Kraft [4, 5], Radecki [12, 13, 14], and others).”
BUELL, e.g., at 61, “We have seen that the usual performance
evaluation measures of recall, precision, fallout, and generality have
analogues in retrieval environments in which decisions
about "retrieval" and "relevance" are no longer Boolean. Although
problems do exist in interpreting the numerical values which will be
obtained from these measures and in comparing those values to those
obtained from Boolean retrieval systems, the problems can be
overcome by taking into account the nature of the indexing function.
Finally, we have raised the question as to whether rank-order
comparison measures might not be more appropriate for evaluating
those systems whose natural output consists of rank orderings of
documents.”

Bill Buckles, “An
Information Retrieval
Perspective on Fuzzy
Database Systems,”
Advances in
Information Retrieval,
ACM 82 Panel Session
(“BUCKLES”)

SALTON, e.g., Abstract, “In conventional information retrieval
Boolean combinations of index terms are used to formulate the users'
information request. Boolean queries are difficult to generate and the
retrieved items are not presented to the user in any useful order. A
new flexible retrieval system is described which makes it possible to
relax the strict conditions of Boolean query logic thereby retrieving
useful items that are rejected in a conventional retrieval situation. The
query structure inherent in the Boolean system is preserved, while at
the same time weighted terms may be incorporated into both queries
and stored documents; the retrieved output can also be ranked in strict
similarity order with the user queries. A conventional retrieval system
can be modified to make use of the flexible metric system. Laboratory
tests indicate that the extended system produces better retrieval output
than conventional Boolean or vector processing system's.”
BUCKLES, e.g., Abstract, “Database in which domain values are not
crisp and precise exhibit properties normally associated with
information retrieval systems. For instance, a boolean query induces a
membership value for each tuple (i.e., record) that is analogous in
function to a similarity measure. Thus, precision and recall measures
are legitimate areas of interest that pertain to fuzzy databases but not
ordinary databases. These ideas will be expounded in the context of a
database for expert advice on national energy policies.”

Donald Kraft,
“Generalizations of
Boolean Query
Processing,” Advances
in Information
Retrieval, ACM 82
Panel Session (“KRAFT

KRAFT II, e.g., Abstract, “Substantial work has been done recently
applying fuzzy subset theory to the problems of document and query
representation and processing in retrieval systems. The motivation
has often been to generalize Boolean query processing to allow for
non-Boolean index weights or measures of importance to be attached
to the individual terms in the document or in the query representation.
The problems of generalizing the Boolean lattice structure have been

Gerard Salton,
“Extended Boolean
Information System,”
Advances in
Information Retrieval,
ACM 82 Panel Session
(“SALTON”)

307

Reference
II”)

Disclosure
noted. Criteria have been generated for query processing mechanisms
with relevance weights in the query, but these have been shown to be
inconsistent. An alternative approach using thresholds in the query
has been suggested, with the generation of appropriate document
evaluation criteria for Boolean query processing.
Problems remain unsolved. The exact form of the function to be used
for the query processing mechanisms must still be specified and
appropriate parameters must be obtained. Some researchers still
prefer a vector space approach, others suggest alternatives to Boolean
queries, others work on probabilistic approaches, and still others
propose new lattice structures for weighted retrieval. These various
models must he reconciled with each other and with an overall
generalization that encompasses each and allows for analysis and
comparison. Moreover, evaluation mechanisms must be sought for
fuzzy systems, and it is necessary to generate a fuzzy concept to the
notion of "retrieval" itself.”

George Baklarz, “Using
Neural Nets to
Optimize Retrieval in a
Fuzzy Relational
Database”
(“BAKLARZ”)

BAKLARZ, e.g., Abstract, “This paper examines the theory behind
Fuzzy Sets and Back-Propagation Neural Nets, and how neural nets
can be used to replace fuzzy sets and improve the query performance
in a Fuzzy Relational Database (FRDB).”
BAKLARZ, e.g., at 191, “In most database systems, information is
assumed to be exact, correct, well formulated , with no provisions for
considering otherwise[l] . Because fuzzy set theory gives us a basis to
manipulate real-world data in a formal way, this technology can be
adapted to relational databases . By extending fuzzy set theory to
relational databases, the user has the added benefit of:
• Not having to state precisely the attributes of a query
• The data can be represented in a fuzzy state
• The relationships can be tailored to the individual user”
BAKLARZ, e.g., at 192, “This paper examines how fuzzy-set theory can
be used in a relational database to better model the information and
facts available to the user. Although there have been various
implementations of Fuzzy Relational Data - bases, the implementation
described here optimizes information retrieval by using neural nets as
a replacement for relations.”
BAKLARZ, e.g., at 200, “Merging fuzzy-set theory with database
technology is a powerful tool for manipulating imprecise information.
By combining fuzzy set extensions to Structured Query Language
(SQL) statements, a user can retrieve data based on imprecise
information . Finally, introducing neural nets as a replacement for
308

Reference

Disclosure
membership functions, dramatically reduces the retrieval speed and
further enhances the usefulness of a fuzzy relational database. The
existing prototype has highlighted a number of areas that warrant
further research :
• Neural net selection
The back-propagation neural net was chosen due to its storage
capacity and for its ability to learn a variety of equations, but other
neural net models may be more suitable.
• Training Techniques
The training time of the neural net needs to be improved.
• Parallel Processing
The back-propagation algorithm is well suited to parallel
implementations, and implementing the algorithms on a parallel
machine will highlight the performance benefits of using neural nets.
Neural nets have proven to be a very useful replacement for relations
in a Fuzzy Relational Database. Fuzzy set theory and neural nets
complement one another, and this knowledge should lead to more
applications where neural nets can replace fuzzy membership
functions to improve performance.”

P. Subtil et al., “A
Fuzzy Information
Retrieval and
Management System
and Its Applications,”
(1996) (“SUBTIL”)

SUBTIL, e.g., Abstract, “This paper presents a fuzzy information
retrieval and management systems (FIRMS) we have developped for
handling fuzzy objects. The originalities of this system consist of : i)
the possibility to describe object with fuzzy aggregate attributes and to
retrieve them at different description-levels of these attributes, ii) the
definition of nuanced domain which gives the possible values of a
fuzzy attribute, iii) the using of a fuzzy thesaurus and an associated
grammar to go through its links in order to retrieve objects. In another
hand, this paper explains how to build an application with this system
and shows some real applications of FIRMS.”
SUBTIL, e.g., at 537, “Vagueness and uncertainty are usual in the
human knowledge and reasoning. Then it is necessary to handle these
fuzzy data in databases when they are the only information known
about the world to model. During the last years, several approaches
(see [9] and [3] for example) have proposed extension of databases to
take into account this imperfection of real world. Allmost of this
approaches use the concept of fuzzy set [10] and possibility theory
[11]. In this paper, we present our approach about a fuzzy information
retrieval and management system called FIRMS. In the first section,
we present the basic concepts of our system whence the original
concepts of aggregate attribute and fuzzy thesaurus. In the second
section, we present the modelling process of an application, an
application being a set of fuzzy object. In the last section, we present
real applications of FIRMS in economic fields.”
309

Reference

Disclosure
SUBTIL, e.g., at 537, “FIRMS allows to describe a set of objects
defined by a collection of attributes. For instance, John is an object
defined by the attributes Name, age, . . . . An attribute can take a fuzzy
value called a nuanced value. In the first subsection, we give th three
kinds of attributes used by FIRMS. In the second, we explain the
concept of nuanced domaingrouping the possible values of an
attribute. In the third subsection, we introduce the notion of fuzzy
thesaurus as a particular nuanced domain. In the last subsection, we
show the description of a fuzzy object.”
SUBTIL, e.g., at 540, “Comnent of an european research system, the
panel of Lorraine PME-PMI (Lorraine is a kind of state in France and
PME-PMI designates firms under 500 employees) is a very important
piece of a program developed by Institut Commercial de Nancy
(commercial institute of Nancy) and the Conseil Rdgional de
Lorraine. This panel must allow to know the management method of
firms more precisely. It groups data on more 400 firms from 1989 to
1993. Each firms is defined by 700 variables about the following
subjects : product, rivalry, export, human resources, strong and weak
points, innovation, national assistance, ...Among these themes, we
have selected those which present some interest for the representation
of vague and/or uncertain information : strong and weak points,
export, human resources and performance. We use the system FIRMS
essentialy for two reasons. Firstly, the system allows a reality
representation more reliable with the inherent vagueness and
uncertainty. Secondly, the system allows rapidly and simply
verification of hypothesis by the use of profiles.”
SUBTIL, e.g., at 540-41, “The Institut Commercial de Nancy has
developed an expert system in 1989 to formulate dignosis about the
statement and the development of a firm. This analysis uses dynamic
contextual factors which continually have an influence on firms
framework. But the expert, using a set of simple rules, was not
efficient because it cannot handle vagueness. For instance, it cannot
take into account the following rules determined by an human expert
to qualify an emergent firm.
If
roduction cost = high) and (customs experience
= low) and (existence of infrastructure
= low) and (strategic uncertainty = high)
then emergent firm
FIRMS has been used to solve the problem of vagueness and
uncertainty. A firm is described by a list of attributes : production
cost, customs experience, existence of infrastructure, strategic
310

Reference

Disclosure
uncertainty, technological resources, technological uncertainty, . . . .
All attributes has the same nuanced domain.”
SUBTIL, e.g., at 541, “FIRMS is a flexible system allowing the
handling of vagueness and uncertainty. Among the basic concepts,
two are very important. The first is the concept of aggregate attribute.
It allows an user to describe two objects with different precision
levels. The second is the concept of fuzzy thesaurus. It determinates a
set of weighted linked terms and uses a grammar to go through it
contrary to other approaches [4][5]. An iterative and incremental
process allows to describe the basic elements of an application. The
experiences with real data in economic fields has shown the flexibility
of FIRMS.”

C.T. Yu, “An Approach
to Probabalistic
Retrieval,” (1981)
(“YU”)

YU, e.g., Abstract, “The objective is to relate the effectiveness of
retrieval, the fuzzy set concept and the processing of Boolean query.
The use of a probabilistic retrieval scheme is motivated. It is found
that there is a correspondence between probabilistic retrieval schmes
and fuzzy sets. A fuzzy set corresponding to a potentially optimal
probabilistic retrieval scheme is obtained. Then the retrieval scheme
for the fuzzy set is constructed.”
YU, e.g., at 46, “The effect of term weights on the performance of
queries was analyzed in [24] , where it was shown that queries whose
terms having higher "precision values" are assigned heavier weights
yield better retrieval results than queries whose terms are assigned the
same weights, under the assumption that terms are distributed
independently. Thus, the precision value of a term characterizes the
usefulness of the term in retrieval. This result was supported in [13],
whre it was shown that if terms of a query are assigned weights
proportional to the logarithm of their precision values, then optimal
retrieval results are obtained under the same term independence
assumption. When terms are distributed dependently, the
incorporation of the term dependence into the retrieval process yields
better retrieval results [9,20,23]. Even more general condition exists
for the construction of the optimal queries [5,8,21]. The above results
assume that certain parameter values (e.g. those needed to compute the
term precision values) are known. When these values are not known,
they may be estimated by relevance feedback [5,7,15,22] where the
user identifies each retrieved document as either relevant or irrelevant,
and input the information to the system. Where relevance feed back
can not be employed (e.g. a user submits a query the f i r s t time),
various attempts have been made [6,17,18,25] to yield reasonable
retrieval results. All these techniques are used when the user's queries
are expressed as sets of keywords.”
311

Reference

Disclosure
YU, e.g., at 46, “The use of a probabilistic retrieval scheme (PRS) is
motivated. It is applied to the processing of Boolean queries. Our aim
is to obtain a potentially optimal PRS. To achieve this, a
correspondence between PRS and fuzzy sets is established. A process
to obtain a fuzzy set corresponding to a potentially optimal PRS is
presented. Then, a potentially optimal PRS is constructed from the
fuzzy set. Finally, the performances of some natural retrieval schemes
are compared using a partial ordering deduced from a given Boolean
query.
The main contributions of the work presented here are (1) a
relationship between a retrieval scheme and its retrieval effectiveness
is established analytically;
(2) the use of fuzzy set, which has been employed by earlier
researchers but not related to the effectiveness of retrieval, fits into the
development of (1) naturally; and
(3) a conceptually very simple process to obtain a potentially optimal
PRS is provided. This procedure is independent of the given partial
ordering. Thus, if a better partial ordering (than the one given here) is
obtained by another interpretation of a Boolean query or by re-evance
feedback, the procedure given here can still be applied.”

Gary Mooney,
“Intelligent information
retrieval from the
World Wide Web using
fuzzy user modelling,”
Library and Information
Research News, Vol.
21, No. 67 (1996)
(“MOONEY”)

MOONEY, p. 25 – “This article investigates the effects of applying
fuzzy logic and user modelling techniques to the process of
information retrieval from the WWW, a major part of the Internet.
This is a novel AI approach to the process of IR. To perform the
investigation, a prototype system, the Fuzzy Query Modelling
Assistant (FMQA), has been developed. The focus of the investigation
was whether the results achieved by using the FMQA would improve
upon those returned by using an existing search tool, specifically
LycosTM (Mauldin, 1996), alone. To answer this question a user
study of the FMQA is being performed and its early results are
reported.”
MOONEY, p. 25 – “A major problem with IR lies in the vagueness and
lack of precision of the prospective searcher's information need. This
vagueness and lack of precision leads to the aforementioned problems
and these are exacerbated by the nature of the WWW. The problem of
information overload is one example. A search with the tool LycosTM
using the search string 'information retrieval' produced 61 ,000+ hits
(Mauldin, 1996). However, information about the user's experiences
and knowledge of the search subject and of the WWW in general can
be used to modify the query intelligently and produce better IR results.

312

Reference

Disclosure
User modelling research has shown that adaptive user stereotypes are
often used to represent different sorts of user and their characteristics
(Rich, 1979). Fuzzy logic, with its inherent ability to capture and
represent partial know ledge, is a valid AI technique to use in IR-a
process involving the representation of information needs as queries,
with all the attendant vagueness and semantic ambiguities (Zadeh,
1993). Here, the user stereotypes are represented as fuzzy sets to
ensure flexibility and adaptivity. This concept is at the heart of the
prototype system, the FMQA.”
MOONEY, p. 25 – “The FMQA is not designed to act as a new
'intelligent' search engine. Within the field of distributed AI and
computing in general there has been much research and development
into the notion of searching and intelligence through the development
of agents.(Wooldridge & Jennings, 1995). The
FMQA seeks to alleviate IR problems through 'intelligently' assisting
the user in a search. In this sense it is similar to the concept of
interface agents defined in Maes(1994) but the FMQA applies fuzzy
logic and user modelling to the query formulation of searches. The
aim is to refine a query before it is submitted to an existing search
engine. This refinement is based on knowledge about the user's beliefs
and experiences (in the Internet and the subject domain of AI)
captured through an on-line interactive session.”
MOONEY, p. 25-26 – “The captured user knowledge is used to adapt
default user models in order to represent an individual user. This
representation is then combined with the user's query to produce the
refined query. The knowledge is captured from two on-line interactive
questionnaires. Each question is represented by a fuzzy set. The FKB
combines the answers to these questions to produce two sets which
represent the individual user model. The defuzzified values from these
sets are used to refine the query.
The user is then given the option to submit either the original or the
refined query to LycosTM. Presently, the user must choose the
original query from a list of topics representing different areas of AI
but this is just a facet of the prototype. The system could easily be
expanded to include other fields of interest and eventually to allow the
user to enter the query words themselves. Additionally, the final fuzzy
sets are lost when a user finishes accessing the FMQA and new sets
are produced with every new session. However, the system could
easily retain these sets and use the information they contain in future
sessions.”
MOONEY, p. 25-26 – “A user study has been performed in which a
313

Reference

Disclosure
number of DMU undergraduate students from the first
year intake of two courses accessed the FMQA and provided feedback
on its use and performance. . . .The study took place over a number of
weeks with each user accessing the FMQA at a similar time each day.
This ensured that each user's reaction to the system was not unduly
influenced by differences in the Internet network traffic speeds. Each
user was asked to submit the original topic they chose as a query and
the refined query produced by the system. This is equivalent to using
Lycos alone to search for the AI topic and then using the same
FMQA-modified AI topic and allows the results to be used to answer
the central question of the study. As part of each session, the user was
presented with an e-mail form and asked to list the best and worst
results for each query, and to rank these as well by giving them a score
between 1 and 10. They were also asked to comment upon the results,
in terms of usefulness and relevance, and upon the system, in terms of
ease of use and design.”

MOONEY, p. 25-26 – “Early results form the study indicate that the
FMQA does indeed improve upon the IR results achieved by using
LycosTM alone. . . . During a query looking for 'Fuzzy Logic', which
afterwards the user remarked that the modifed
results were more relevant, the best result for the modfied query was a
website dedicated to fuzzy sets and systems (Brown, 1996). The
dedicated website contains many WWW links to general sites
of interest to fuzzy logic reserachers and would be a good starting
point for a novice to the area, which was the category this user was
placed in by the FMQA.”
MOONEY, p. 25-26 – “This article has examined the application of
fuzzy logic and user modelling to the process of IR from the WWW,
the concept being to assist intelligently the user in searching for
information and reduce the problems commonly associated with IR in
general, eg irrelevance and redundancy. A prototype system, the
FMQA, was developed, which realises the concept by employing
knowledge about the user to modify queries before they are submitted
to an existing WWW search tool. This knowledge is represented in
fuzzy sets which act as adaptive user stereotypes. Early results from a
live user study of the FMQA show that, in the opinion of the users, the
results achieved from using the system do improve upon those
obtained from using the search tool alone.”

Henrik Larsen and
Ronald Yager, “Query

LARSEN, p. 1 - “The ascendancy of the Internet, and in particular the
World Wide Web, is making the development of intelligent
314

Reference
Fuzzification for
Internet Information
Retrieval,” (1996)
(“LARSEN”)

Disclosure
information retrieval an extremely important issuer. An information
retrieval system[1] is a system to retrieve relevant information objects
from an information base. The information base stores a collection of
objects some of which are of potential interest to the users. Each
object is represented by an item which can be seen to be made up of
two components. The first component is the index and the second
component is the body. The index usually consists of highly organized
pieces of information that can be used to help identify and select the
objects that may be relevant to a user. The body consists of
information which may not be organized but it contains the material
that is of interest to the user. The fundamental problem in information
retrieval is to find the subset of objects in the information base that is
relevant to a given user. In a fuzzy information retrieval system, one
can supply the list of relevant items with an ordering as to their
potential interest to the user. Figure 1 shows a top-level view of the
information retrieval system processes.
In the first step the user enters a request in terms of features of interest
employing the keywords in the indexing system used to describe the
objects. The information in this query is then used by the information
retrieval system to select items that may be potentially relevant to the
user. The final step is a process where the user looks at the items
suggested by the system and decides
the ultimate relevance of the items. This final step greatly reduces the
burden of the information retrieval process, for it allows the user to
look at the items selected and decide the ultimate relevance. This
means that not all the knowledge about the decision has to be
formalized in a manner that can be manipulated by the computer. The
user must only supply the information that is used to search through
the index.
As an example, we will consider the problem of selecting a house for
purchase and assume that the user has access to an information base
consisting of a collection of houses for sale. Here the user would
express desired properties about the kind of house desired (price, size,
location, etc.) in the query. The system would then search the
information base and produce a listing of houses
which closely match the user's request. This information could include
text, more detailed information about the house as well as perhaps a
picture of the house. The user then looks at this information and then
decides which houses he wants to visit. In making this decision, the
user may use all kinds of subjective criteria which may be hard to
quantify and not necessarily specified in his query.”
LARSEN, p. 2-3 – “In this paper we shall describe an information
315

Reference

Disclosure
retrieval system which uses fuzzy sets to help in the selection process,
this kind of system can be viewed as an intelligent inquiry system.
Figure 2, which is an expansion of the information retrieval system
box of Figure 1, illustrates the steps involved in the information
retrieval process.
In the first step the crisp information provided by the user is softened
with the aid of fuzzy sets. Using the index and a modified version of
the requirements (“crisp envelope”, step 2), we search through the
information base (step 3), to find a subset of objects in the information
base that can be considered as potentially relevant to the user. Step 3
can be based on an ordinary crisp querying language. The set of
objects found in this step is called the “crisp
envelope” answer. The final step in the process is a ranking of the
elements in this crisp envelope which is then presented to the user.”

LARSEN, p. 4 – “An important characteristic of many of the criteria
supplied in a user query is that the needs they intend to represent are
not crisp. If persons looking for a house indicate their desire to spend
between $100,000 and $140,000 for the house, it is not the case that
they will be totally uninterested in a house costing $145,000. They
may be less satisfied but not completely unsatisfied. The central
observation here is that the boundary between a criteria being
completely satisfied and not being satisfied is fuzzy rather than crisp.
In building intelligent information retrieval systems we must take
advantage of this fuzziness in the criteria. As we shall subsequently
see, we use this fuzziness in two ways. First, we use it to soften the
user query to allow potential interesting items to be retrieved, even if
they do not directly satisfy the original user query. In particular, we
shall use it in providing a query envelope, that is, a crisp query applied
to retrieve the potentially most interesting items from the information
base. The second way we shall use this fuzzy characteristic is to
provide an ordering (ranking) of the items according to the degree to
which they satisfy the
softened user query.”
LARSEN, p. 5 – “While many of the criteria in a user query can be
softened (fuzzified) with the aid of fuzzy subsets some criteria are not
amenable to this kind of softening. For example, the desire to have a
fireplace or two bathrooms is not easily fuzzified.”
LARSEN, p. 17 - In the preceding, we discussed the issue of criteria
aggregation. We shall now specialize this to the ranking of objects for
an information retrieval system. We discussed two classes of
316

Reference

Disclosure
aggregation, MOM and MAM operators. We recall that the MOM
operator is a generalized or-like aggregation while the MAM operator
is a generalized and-like operator. In information retrieval systems we
see the criteria specified by the user as being connected by an and-like
operator, assuming the user generally wants all the criteria satisfied.
That is, a person desires to obtain further information about houses in
a certain price range and in
a particular location and having certain amenities. Thus, the
appropriate family of operators are the MAM operators.”
LARSEN, p. 20 – “We presented an approach to a weighted multicriteria information retrieval system that uses fuzzy subsets as
mechanism to allow for the flexible evaluation of user requirements.
Although we focused on numerical criteria, the approach is also
applicable for non-numerical criteria (concepts, terms)—in the first
case, the semantic similarity utilized relies on the numerical scale, in
the second case, it relies on a similarity relation. We discussed the
potential use of MAM and MOM operators as a tool for the
aggregation of user requirements. Finally, we illustrated the
application of the mechanism and tools in an application for a real
estate agency. Our an approach is in particular interesting for retrieval
through the Internet WWW. In this situation, the semantic elasticity
supported by our approach allows the user to retrieve the most
interesting objects, even when the description applied in the
information base does not directly match the query formulation chosen
by the user.”
LARSEN, Figures 1, 2, 3

Tadeusz Radecki,
“Fuzzy Set Theoretical
Approach to Document
Retrieval” Information
Processing &
Management, Vol. 15,
pp. 247-259 (1979)
(“RADECKI”)

RADECKI at Abstract - “The aim of a document retrieval system is to
issue documents which contain the information needed by a given user
of an information system. The process of retrieving documents in
response to a given query is carried out by means of the search
patterns of these documents and the query. It is thus clear that the
quality of this process, i.e. the pertinence of the information system
response to the information need of a given user depends on the
degree of accuracy in which document and query contents are
represented by their search patterns. It seems obvious that the
weighting of descriptors entering document search patterns improves
the quality of the document retrieval process.
A mathematical apparatus which takes into consideration, in a natural
manner, the fact that the grades of importance of the descriptors in
document search patterns are of the continuum type, that is an
apparatus adequate to the description of a retrieval system of
317

Reference

Disclosure
documents indexed by weighted descriptors is-among known
mathematical methods-the theory of fuzzy sets, formulated by L. A.
Zadeh.
It is the aim of this paper to present a new method of document
retrieval based on the fundamental operations of the fuzzy set theory.
We start by introducing basic notions, then the syntax and semantics
of the proposed language for document retrieval will be given and an
algorithm allocating documents to particular queries will be described
and its properties discussed.
The basic advantage of the use of the fuzzy set theory for document
retrieval system description is that it takes into consideration, in a
simple way, the differentiation of the importance of descriptors in
document search patterns and the differentiation of the formal
relevance grades of particular documents of an information system to
a given query.
Documents of the highest grades (in the given information system) of
formal relevance to the given query may be retrieved by means of the
application of simple operations of the fuzzy set theory.”
RADECKI, p. 2 - “Of the known mathematical methods, the method
best fulfilling the postulates formulated above, and therefore adequate
for an analysis of document retrieval systems is the theory of fuzzy
sets, whose bases L. A. Zadeh has given in [12-151. The idea of the
theory of fuzzy sets is that the grades of membership of particular
elements of the universe in a given fuzzy set are determined by the socalled membership function which is a generalization of the
characteristic function. The transition from membership to nonmembership of the universe elements in the fuzzy set, in contrast to
the ordinary set theory, is continuous.
Many papers have already been written on investigations into the
possibility of creating a uniform document retrieval system theory
based on the theory of fuzzy sets. Besides the present author[16-201
many other specialists have also dealt with this question[21-271. In
paper[21] C. V. Negoita used the theorem on the separation of fuzzy
sets[l2] to divide a set of document search patterns into clusters where
each cluster is made up of those document search patterns whose
grades of membership in that particular cluster are not smaller than the
established threshold value. The idea of using the theory of fuzzy sets,
or, to be precise, the concept of the similarity relation to formulate an
algorithm for the division of a set of document search patterns into
clusters has also been used in[lS, 22, 25’1. In[22, 25]-making direct
use of
318

Reference

Chris Buckley,
“Implementation of the
SMART Information
Retrieval System,”
Department of
Computer Science,

Disclosure
the definition of the max-min composition[l3] of fuzzy relations- a
way is suggested of dividing the set of document search patterns into
clusters where each cluster is made up of those document search
patterns whose grades of similarity are not smaller than the established
threshold value. One disadvantage of the way of organizing the
document file, suggested in these papers, is that in the case of a large
set of documents the process of dividing the set of document search
patterns into clusters is very time-consuming and also expensive. This
inconvenience can be significantly attenuated by using the method of
organizing the document file proposed in paper [ 181 based on the
notion of the maximum spanning tree. In paper [23] as in paper[24] C.
V. Negoita defines the response of an information retrieval system as
a fuzzy set
and describes the relationships between various responses of the
system in terms of the theory of fuzzy sets. Retrieval methods of
documents indexed by weighted descriptors, which are a
natural generalization of the set theory methods, have been described
in papers[l6, 17, 19, 201 by the author. In paper[26] W. M. Sachs
draws attention to the possibility of defining associative retrieval in
terms of the fuzzy set theory, but does not provide any new solutions
however. On the other hand, paper [27] by V. Tahani, based on an
idea similar to that expressed by the author in paper[l6], contains a
description of the organization of document file and a strategy for the
retrieval of documents using basic notions and operations of the
theory of fuzzy sets.
The aim of the present paper is to describe a generalized method (in
comparison to the strategies presented in papers [ 16, 271) of
document retrieval. In the writing of this paper ideas contained in
previous papers by the author[ 16, 17, 19,20,28] were utilized. Before
entering a detailed description of the proposed method of document
retrieval, we will present the basic notions used in the rest of the
paper. We will then describe the proposed document retrieval
language and present an algorithm for the allocation of documents to
particular queries and describe the properties of the language and the
algorithm. The proposed document retrieval
strategy will also be illustrated by an example. Finally the results of
the present paper will be summarized and modifications to the
document retrieval method presented will be discussed.”
BUCKLEY, p. 2, “The SMART information retrieval package is a set of
programs composing a fully automatic document retrieval system. It
allows easy creation, maintenance, and use of on-line document
collections. As more information is being kept on-line every day; it
becomes more essential to have methods of easy, natural access to the
information. The SMART package is primarily a tool for investigating
319

Reference
Cornell University
(May 1985)
(“BUCKLEY”)

Disclosure
some of these methods. In addition, it is quite usable itself for many
applications.”
BUCKLEY, p. 2-3, “This implementation of SMART contains few new
or radical concepts. Instead, it attempts to provide a solid framework
for future work in information retrieval. The two major goals of the
current version are to
1. Provide a flexible experimental system for research in information
retrieval. See [6] for a discussion of desirable system capabilities and
design principles for experimental work.
2. Provide a fast, portable, interactive environment for actual users.
These two goals naturally conflict with each other; the current
SMART design is an attempt to satisfy each as much as possible. The
system is concerned with three major types of users: the
experimenters, the database administrators, and the naive users. The
experimenters need the ability to easily change system parameters and
to easily add or replace program modules. The database administrators
must be able to create and maintain a collection of documents without
worrying about the peculiarities of the particular. collection. It should
be possible to initially specify the features of the collection and not
worry about them again. The users need to be able to enter a query and
view the results without knowing anything about the internal
parameters of the system, being aware only of the collection features
which are relevant to them such as the type of information contained
in a document). An interactive help facility is necessary for the casual
user. The current system is a first step in satisfying these goals. The
major lack at the moment is a satisfactory user interface. There is a
usable interface here at Cornell, but more work is needed.”
BUCKLEY, p. 3, “The design of the SMART system concentrates on
two types of flexibility. The first is complete flexibility at a number of
levels in specifying the parameters for all operations. All parameters
have reasonable default values. In addition they (possibly) can be
given values within a collection dependent specification file. This
means a database administrator can tailor the parameters to one
particular database application. These values, in turn, can be overridden at command execution time by specifying a parameter and its
value on the command line. At the program design level, flexibility is
achieved by allowing very easy expansion of the most commonly used
modules. For example, if an experimenter wishes to add a new
procedure for computing the similarity between two vectors, two lines
in one "data" file needs to be changed and the retrieval program needs
to be re-linked.”
BUCKLEY, p. 5-6, “Users come to the SMART system with an
320

Reference

Disclosure
information need and try to convey this need to the system. Their
initial statement of their need can be a piece of natural language text, a
query using Boolean connectives (AND, OR), a list of keywords, etc.
The system assigns a query representative for the need, either a simple
list of concepts and weights like the document representatives, or
something a bit more involved which gives more structure to the
representative. A retrieval function within the system then calculates
the similarity of the query representative to each of the document
representatives. (In practice, not every document needs to be
examined - depending on the similarity function.) The documents are
presented to the user in order of their similarity to the query. It is
hoped that the similarity order will have some correspondence to
likelihood that the user will judge the document useful.At this point,
the user has the option to examine some of the top retrieved
documents, and give a judgement of whether the documents were
relevant to their information need. If the user desires more documents,
a new query representative can be automatically constructed from the
old representative and some of the concepts occurring in the relevant
documents. This process is known as relevance feedback. The new
feedback query can then be compared against the document collection
and more documents can be retrieved for the user. This process
continues until the user has as many documents as they desire.”
BUCKLEY, p. 13, “There is only one program, retrieve, in the retrieval
module, but it is a very
flexible program! Retrieve runs an indexed query collection (possibly
consisting of just one query) against an indexed document collection,
calculating (theoretically) the similarity between each document and
each query. The output is either a list of the documents which most
closely match each query or a list of a given set of documents and the
ranks which would be assigned them if the documents were sorted in
decreasing order of similarity to the query. In an experimental
research setting, this set of documents would be the known relevant
documents for each query and The ranks of these relevant documents
are used to evaluate the
effectiveness of different retrieval methods. All of the options of
retrieve are given in the retrieval specification file passed to it. These
options include information like
1. Type of input query (vector, boolean tree, pnorm)
2. Retrieval method to be used (discussed below)
3. Type of output desired (just top documents, ranks of relevant
documents,
both)
4. The location of the input (document collection, query) and the
output.
5. Etc.
321

Reference

Disclosure
The parameters whose values can be specified within the specification
file are given reasonable default values. For most operational runs, as
opposed to experimental runs, the specification file consists of a single
line telling what collection is to be used. On the other hand, a
complicated experimental run that, say, uses a different matching
function for every type of information in the query, could run to 30
lines of parameters. The various retrieval methods form the heart of
retrieve. To allow complete flexibility, there are three levels of
retrieval methods: the collection access level, the vector access level,
and the ctype access level.”
BUCKLEY, e.g., p. 26, “Two methods are defined for accessing a
dictionary entry: hashing on <token,ctype> or direct access through
<con>. <con> is simply the dictionary entry index that <token,ctype>
hashes into when the entry is originally placed in the dictionary. Thus,
a quick direct access to the token and freq values exists given the
values of <con>. This is used (possibly) during retrieval and feedback
operations. There may be some similarity computations based upon
the token for example, experiments using fuzzy matching of dates),
and the freq information is used extensively by feedback. Accessing
via <token,ctype> is essential during the indexing process.”
BUCKLEY, e.g., pp. 35-36, “There is very little that is new about the
current design of SMART. Instead, the standard information retrieval
algorithms are implemented in an efficient and flexible manner. The
core of the system is the set of low-level data access mechanisms that
allow the rest of the system to look at stored information as sequences
of tuples and to efficiently access individual tuples. The experimenter
and database administrator are aided by a uniform approach to
specifying parameter values. A rudimentary user interface exists that
allows interactive help for many purposes. Concurrency issues in
SMART are dealt with superficially, but in a manner that should be
sufficient for most non-commercial uses of the system. The resulting
system turns out to be quite usable for both casual and experimental
purposes. A casual user can submit a query and receive back the
relevant documents within a couple of seconds. The experimenter can
change parameters and even algorithms with minimal effort. For
example, one recent investigation into term weighting schemes
involved implementing several different term weighting methods. It
took 1 day (about 25 hours) to implement, run, and evaluate the
methods (a total of 119 experimental runs were made). This type of
investigation would previously have taken a couple of weeks. There
are still a number of problems with SMART. The foremost of these is
the user interface. There are clear improvements that can be made in
the present interface; the need for other improvements will become
322

Reference

Disclosure
obvious as the system is used by more people. Another area for
improvement already discussed is that of concurrency. Both the user
interface and concurrency problems stem from the gradual change of
SMART from an entirely experimental system to one that can be
actually used. A number of the algorithms used in the implementation
can be improved. In general, straight-forward algorithms were
preferred. More complicated algorithms which are more efficient,
especially space efficient, exist and should be implemented. The
dictionary access procedures are a good example of this. The number
of applications for SMART will undoubtedly increase in the next
couple of years. At this time at Cornell, it is being used for
1. Searching a collection of CACM abstracts
2. Providing a help facility for UNIX. There was a lot of
documentation for
UNIX on-line that was inaccessible because nobody could find it.
3. Accessing a user information database (interests and hobbies as
well as
factual information).
4. Accessing reference databases (easy, non-factual searches of
standard databases
of references)
5. Searching electronic mail files (eg. the old mail to system support
staff)
6. Searching archives of electronic bulletin boards (USENET news)”

NAQVI WO

NAQVI WO, p. 5-6 - “When the user requests a certain page or a
certain topic of information, the relevant pages are retrieved from
the computer network and shown to the user. The present invention,
upon receiving the user's request, retrieves advertisements that are
related to the user's action, dynamically mixes the advertisements with
the content of the pages according to a particular layout, and displays
the pages with focused, targeted advertisements as a part of the page.
The advertisements can be made to satisfy a set of constraints
requested by the advertiser, as well as the constraints of the publisher
of the page, as further discussed below. The advertisement triggering
mechanism of the present invention is not random or coincidental, but
rather, is prespecified in advance. This specification will be referred to
in this application as a contract. A contract specifies the marketing
rules that link advertisements with specific queries. For example, a
diet soft drink advertisement may be shown when a user asks for a
page about exercising equipment. These rules are specified by
advertisers implementing the concept of "focus" or "relevance" of
advertisements and help the advertisers to target a specific audience.
Owners of pages specify the focus content of their pages through
special tags within a page. These tags are not displayed to the
323

Reference

Disclosure
information consumer; the tags are used to decide what advertisement
can be shown when the page is requested by a consumer. The notion
of a contract, however, goes well beyond
just marketing rules. First of all, the advertising space
on the online medium, although technically unlimited, is
severely restricted by the user's attention span. Placing advertisements
on the first page which constitutes the answer to a query gives the
advertisements much higher probability to be seen than on later pages
of the answer.”
NAQVI WO, p. 15-16 – “Initially, a user requests a particular piece of
information through one of the clients 17. The user's
10 request is given to the WWW Daemon 16, which passes the
information to the gate 15. The gate 15 at this point
decides what piece of information is being requested by the
user and finds other relevant pieces of information that
can be commingled with what the user has asked. The user,
15 for example, might ask the system to see certain car
dealers, to find a phone number of a car dealer, or to get
a page of a particular magazine. The gate 15 at this point gives the
request to the matching rule engine 18 ("MRE"). The purpose of the
MRE 18 20 is to look at the content of the user's query and to find a
category within its active index SIC 19 that matches the
same type. If the user has asked for car dealers, the MRE
18 invokes its rules to determine that car dealers are part
of a class of things relating to transportation. Based on
25 the classification determined by the MRE 18, the system now
knows that the user is asking about cars or about
transportation or about whatever else that the user might
be interested in. The MRE 18 at this point then returns to the gate 15
30 the category index of the user's query. If the user had
asked about cars or about family sedans or about sports
cars, at this point the MRE 18 would have figured out that
the user's interest falls into a certain category. Based
on the user's interest category, the system then retrieves
the advertisements that are relevant to that category.
Thus, the purpose of the MRE 18 is to figure out what the
5 user requested, to place the user's request in a category
of a classification system (i.e., the active index SIC 19)
and, based on that classification, to retrieve relevant
advertisements.”
NAQVI WO, p. 20 – “During the computation of the advertisements
and all the other computations that the system of the present
5 invention performs, a logging module 22 of the system
324

Reference

Disclosure
performs extensive logging of what the user has asked, what
advertisements were shown, how long the advertisements were
shown, and which advertisements were shown to which user.
The logging module 22 then stores these logs in a SYS logs
10 database 23. Various scanned reports can be produced and
defined using the information in the SYS logs database 23.”
NAQVI WO, p. 24-25 - In using a yellow page publisher there are two
broad 20 distinctions for a query. A client may be asking for a
certain category of listings, or the client may be asking
for a particular vendor. For example, the user could ask
for car dealers in Morristown, NJ (i.e., a category of
listings), or the user could ask for Morristown BMW located
25 on South Street in Morristown, NJ (i.e., a particular
vendor) . The system determines which of the two types of
queries or searches the user has made, as illustrated by
box 32 in Fig. 2. If the query is for a certain category,
the process will go to the left hand side of the flow chart
30 of Fig. 2, and if the query is for a certain vendor, the
process will go to the right hand side of the flow chart of
Fig. 2. The left hand side of the flow chart will be
explained first.
After determining the type of query, the category
search engine 33 next determines which category best fits
5 the user's request. The user may have asked for "car," but
the category in the yellow page provider's index may in
fact say "automobile." Or, the user may have asked for
"spectacles," and the category in the yellow page provider
may be called "optician." The matching of these variations
10 of terms is performed by the category search engine 33.”
NAQVI WO, p. 26-27 – “The "focus" arrows 43 shown in Fig. 2
indicate that a certain focus is associated with each category. The
query
may have been directed to a category of listings or a particular vendor.
In both cases there is a "focus"
associated with the content of the query (e.g.,
automobiles, physicians, lawyers, etc.). In addition,
there may be a focus associated with the geographic
5 location of the user to permit advertisers to target users
in particular geographic regions. The focus process plays
a major part in the present invention. No advertisements
are shown unless it can be determined that the
advertisements are in some way focused or related to the
325

Reference

Disclosure
10 content of what the user requested.”
NAQVI WO, p. 34 – “To start (step 80), the user enters a query. For
example, the user may enter restaurants or cars as a query.
The query has a focus, as described above. The system
determines what the focus is and, as described above, the
25 system provides the user with a list of categories that
relate to the query. For example, if the user requests
restaurants, the user might be shown a list of restaurant
types, such as Chinese, American, French, Italian, and so
forth. The query entered by the user is evaluated by a
30 query form manager (step 81) to determine the focus of the
query.”

BULL

Figures 1, 2, 7, 8B, 10, 11 (and associated text)
BULL at Col. 3 – “The user logs on to the system either by name,
address, etc. or with some pseudoonym (or some combination). This
allows the user’s activity to be tracked and establishes a log of the
user’s activity during the current online experience (session). The user
is also asked for explicit profile information concerning preferences.
These preferences will be used to narrow the information retrieval.”
BULL at Col. 5 – “IV. Automated Profile Generation.
Presently, user’s profiles are collected based on explicit
entry by the user, and extraction from demographic data
collected from a variety of sources. In the present invention, the
searching patterns of the user on the Internet are monitored. A set of
software text agent profiles is developed and may be integrated with
explicitly collected profile information. The automated profile
generation will have both explicit profile information gathering and
implicit profile information gathering capabilities.
As the user uses the information aggregation and synthesization
system, the pattern of information being viewed is analyzed and the
user presented with search ideas as well as promotions and specials
from suppliers based on these patterns.”
BULL at Col. 6 – “A theme or definition of a class of information (e.g.,
central California travel and tourism or new automobiles) is
identified. Data sources (Local DataStores (500 . . . N) and
Network Accessible DataStores (300 . . . N)) are screened
for relevance, quality of information and appropriateness (or
may be included de facto based on their title or description).
These are indexed using a text indexing software tool 2981
and the indices stored on the system index DataStore 220.
An initial set of Preestablished Software Text Agents are
326

Reference

Disclosure
defined. These agents are words or combinations of words
that form a word based search pattern. This initial set of
agents is relevant to the searches that might be performed
against the class of information that was indexed. (i.e.,
Agents about automobiles would be developed to search a
class of indexed information about new cars). These are
stored in the Preestablished Software Text Agent DataStore
231. The System 200 uses any multipurpose computer
central processing units with the ability to handle multiple
inputs and outputs with the necessary hard disk storage and
to run World Wide Web (WWW) or other network server
software.”
BULL at Col. 12 – “IV. Automated Profile Generation
Browsing patterns of the user are analyzed and these
patterns update profiles automatically.
FIG. 7 illustrates a how diagram for the Automated Profile
Generation. The looking patterns of the user are monitored to develop
a set of software text agent profiles that are integrated with explicitly
collected profile information to assist the user in
narrowing down information for future sessions as well as
suggesting references, merchandise or services during the
current session. This is accomplished by statistical analysis
of the text stream.
The searching patterns of the user on the Internet are
monitored by monitoring the text stream. A set of software
text agent profiles is developed and may be integrated with
explicitly collected profile information. The explicit infor
mation is gathered by queries to the user. The explicit and
implicit data are merged to develop software text agents that
support the user’s future shopping sessions.”
BULL at Figs. 1 - 7 (and associated text)

KOHDA ’96

KOHDA ’96, §2.2: “Note that the agent is aware of the identity of the
user and which page the user is about to read on the browser, so the
advertising agent can tailor advertisements for individuals and their
current interests. Thus it prevents the user from having
to see advertisements that are unrelated to their current interests.”
Id., §3.1: “At invocation, environment information is passed to each
filter program as invocation parameters. The environment information
includes at least the identity of the user and information about the
selected anchor. The contents of a Web page designated by the anchor
are input into the pipe of filters, and the output from the pipe is
327

Reference

KOHDA ’853

Disclosure
displayed on the browser's window as an HTML document.”
Id., §3.2: “The filter keeps in memory the contact path (URL) to the
agent's Web server. When it is invoked, it forwards the invocation
parameters passed from the browser to the agent's Web server, and
waits for a reply.”
KOHDA ’853 at 38:30-35: “the advertising information server provides
the advertising information automatically based upon the retrieval
condition data, wherein another predetermined tag is added to the
provided condition data to retrieve advertising information, and is
derived from the retrieval information.”
Id. at 23:60 to 24:7: “When the user is obtaining the information about
the sales conditions of the latest automobiles, the information server
100 to obtains and analyzes the retrieval information to be obtained by
the user, and recognizes that the information relates to the sales
conditions of the latest automobiles.... Then, the information server
102 selects the advertising information about, for example, sports cars
from a large volume of advertising information relating to
automobiles, and transmits the selected information to the information
retrieving apparatus 100. As a result, the advertising information in
which the user may be interested can be transmitted to the user,
thereby enhancing the advertising effect.”

Sung Myaeng and
Robert Korfhage,
“Integration of User
Profiles: Models and
Experiments in
Information Retrieval,”
Information Processing
& Management,
Vol. 26, No. 6 (1990)
(“MYAENG”)

MYAENG, Abstract, “One difficult problem in information retrieval
(IR) is the proper interpretation of user queries. It is extremely hard
for users to express their information needs in a specific yet
exhaustive way. In an effort to alleviate this problem, two theoretical
models have been proposed to utilize user characteristics maintained
in the form of a user profile. Although the idea of integrating user
profiles into an IR system is intuitively
appealing, and the models seem viable, no research to date has
established a foundation for the roles of user profiles in such a system.
Aiming at the investigation of the roles of user profiles, therefore, this
study first identifies and extends various query/profile interaction
models to provide a ground upon which the investigation can be
undertaken. From a continuum of models characterized on the basis of
interaction types, metrics, and parameters, nearly 400 models are
chosen to investigate the “model space.” New measures are developed
based on the notion of user satisfaction/frustration. In addition, three
different criteria are used to guide users in making judgments on the
quality of
retrieved items. Analysis of the data obtained from the experiments
shows that, for a wide variety of criteria and metrics, there are always
some query/profile interaction models that outperform the query alone
328

Reference

Disclosure
model. In addition, preferable characteristics for different criteria are
identified in terms of interaction types, parameters, and metrics.”
MYAENG, p. 719, “The problem of retrieving information from natural
language databases has been studied during the past quarter century.
In traditional context, retrospective information retrieval (IR) systems
are those in which a user initiates the search process by means of a set
of active queries and receives a set of references to items of potential
interest. One difficult problem in such systems is the transformation of
the user’s information need to the form of an explicit query which
accurately matches the original intention, and retrieves all items of
interest in the database being searched, and only those. Therefore,
users often have great difficulty in using an IR system successfully
regardless of the query language implementation (e.g., a vector form, a
boolean expression of terms, a combination of both [1,2,3], or other
retrieval models [4,5,6,7,8]). As a result, user queries are not
completely satisfactory in expressing the needs in most retrieval
situations. It seems natural that the output of a system based on such a
query is necessarily incomplete and unsatisfactory.”
MYAENG, p. 720-21, “The difficulty of adequate query formulation
also seems related to the subtlety of the human information seeking
behavior. Widely recognized is the fact that different users usually
expect different sets of items from the same query and make different
relevance judgments on the same retrieved items. This means that user
variability should be considered as a factor in information seeking
process [ 121 and incorporated into the system design in some way.
However, since the typical communication achieved between a user
and a system is only through a set of queries and a set of retrieved
items, this somewhat narrow and restricted channel inhibits the system
from catering to the individual’s variability in terms of information
needs.It is conceivable that by maintaining characteristics of an
individual user in the form of a profile, the bandwidth of the
communication channel can be widened. Used as a way of improving
the level of user/system communication effectiveness, the profile
information
is expected to allow the underlying system to understand users better
and to improve the quality of a retrieval output. For instance, the
profile information allows a different interpretation of a query to
produce a different result, and helps the initial output to be tailored to
the user’s particular needs and ranked appropriately, based on the
user’s preference. While the use of tools such as thesauri and
stemming algorithms for a priori processing of a query aims at better
query interpretation by depersonalizing the query in a sense, profiles
are used for the same purpose by personalizing the query [13].
329

Reference

Disclosure
The influence of the user profile on the quality of output depends on
various factors. One important and immediate consideration is how to
modulate the interaction between a query and a profile, so that
reasonable quality of information is maintained. Some models of
query/profile interaction have been developed and their theoretical
foundations have been established [ 14,15,16]. Another aspect to be
considered is how to maintain user profiles. Assuming that reasonably
well-constructed profiles increase the system effectiveness, the nature
and quality of the information in user profiles should determine the
degree of improvement. Recognizing that people tend to be poor at
self-description, a method of automatically and dynamically updating
user profiles has been proposed to facilitate an intelligent and
personalized IR system [17]. Researchers have recognized directly or
indirectly the need for user modeling in various information systems.
Given that information seeking is part of the problem solving process,
it is difficult to study information seeking apart from a particular
context or process [12]. In particular, IR system outputs need to be
produced based not only on the topicality of documents and queries,
but also on informativeness, often affected by such factors
as novelty, understandability, the order of output presentation, and the
suppression of redundancy [18], which are dependent on individual
users. If an IR system is to be designed to take into account individual
variability in backgrounds, interests, preferences, or other significant
characteristics, it becomes obvious to develop a form of user models
for individuals. Nonetheless, the possibilities for user representations
have been explored only to a limited extent in experimental IR
systems [ 191, and uncertainty about how to incorporate knowledge
about users into system design is a major stumbling block in designing
effective IR systems [20]. Indirect uses of the concept of user
modeling in IR are found in [21] and in [22,23].
This study aims at demonstrating the superiority of IR systems with
profiles, a limited form of user models, to those without profiles, and
investigating the query/profile “model space” in order to develop a
theory. In this paper, we first present the “model Integration of user
profiles 721 space” constructed by identifying and extending the
existing query/profile interaction models and then report the results of
a series of experiments conducted to meet the objectives.”
MYAENG, p. 721, “Since this research aims at investigating the roles of
user profiles in a general sense, various interaction models have been
reviewed and extended to serve as a ground on which
the investigation can be undertaken. Given that a profile contains
information about a user’s (or a group of users’) interest, it may be
used in three distinct ways, depending on when and how it is applied
to the retrieval process. First, the interest profile can play a role in
330

Reference

Disclosure
preprocessing a query to produce a modified query to be used in the
subsequent retrieval process. Second, the profile and the query can be
considered as the same kind of entity directing the retrieval process.
The third possibility is to treat the profile as a filter to postprocess
outputs retrieved based on the query alone. Although each method
possesses its own potential merit, the first two have been the focus of
this research; they lend themselves to the theoretical framework
developed to date.
Even with two methods of using interest profiles, there is a continuum
of models from which 396 different models have been identified and
selected to investigate the “model space. ” For ease of manipulation
and theory development, they are organized along three different
dimensions:
1. modes of query/profile interaction,
2. parameters embedded in the interaction modes, and
3. metrics used to discriminate among documents.”
MYAENG, p. 727, “An experimental retrieval system called PBS
(profile-based system) has been developed for this research. In
addition to common features such as accepting a query, searching a
database, and retrieving document surrogates, it provides capabilities
to handle profiles and evaluate different models based on a query and
a profile. The database consists of 3703 abstracts
of Communications of the ACM from 1958 to 1985. Some standard
methods have been employed to analyze and prepare the database for
the retrieval purpose. For example, a stemming algorithm was used for
both database processing and query processing, and the methods of
computing discrimination values and term frequency information [l]
were adopted to compute weights on term-document pairs. Details of
the structure and components of the PBS as well as methods used for
the database process are found in 1261.
Considering the large number of models being tested, the goai of the
experimental design was to maximize the efficiency of available
human resources and minimize the error variances, especially those
which might be incurred from uncontrolled individual differences. To
this end, every query was processed by all models against the
document database so that systematic differences among queries, and
hence among users, co&d hardly mask the actual differences among
models. The experimental design had to overcome two difficulties. It
is well known that sequencing of the output affects a user’s judgment.
That is,
if document Dz is seen after document Di then the user’s judgment of
D2 is affected by the judgment already made on Di . A similar
sequencing effect pertains across models: judgment of the output of a
given model is affected by the judgement of prior models. To
331

Reference

Disclosure
minimize sequencing effects two strategies were used: the output from
a11 models was merged into a single set, and the documents were
presented to the user in a randomized order rather than in an order
related to their presumed relevance.”
MYAENG, p. 728-29, “After an introductory session [26], the first nontrivial task for a subject was to construct a profile as a list of weighted
terms that represent his or her real-life interests within the discipline
of information and computer science and engineering. The following
is an example of a profile constructed by a subject whose main interest
lies in AI in general and human/computer communication interface in
particular:
((artificial 7) (intelligence 7) (communication 10) (interface 7)
(human 3) (factors 3) (network -2)).
The last term ‘network’ with a weight of -2 was used to explicate her
disinterest in the area of networking, which otherwise might be
implied by the inclusion of the term communication. Unspecified
weights defaulted to a value of 1. Subjects were then asked to
formulate a query to be searched against the database in the PBS.
There was a time interval of at last one day between profile
construction and query formulation, which supposedly reduced any
unnecessary dependence of a query on the content of a profile.
Although they were encouraged to bring their own current information
needs for queries to be submitted to the PBS, a pool of real questions
drawn from comprehensive examinations given by the DIS at
University of Pittsburgh was available as a guide to help them in
conceptualizing and defining an information need and thus a query,
not as a depository from which they should select an information
need. When the subjects were given a randomized list of documents,
they went through documents
in that order and determined the quality of each document based on
three criteriarelevance, pertinence, and usefulness. These fine-grained
criteria were used to forcefully avoid confusion as to how the general
term ‘relevancy’ can be interpreted, as well as to observe what aspects
of ‘relevancy’ are affected by the use of profiles. Relevance was to be
judged objectively based on how closely a document was related to a
‘stated query’, regardless of the user’s expectation or intention.
Pertinence, in contrast, was to be judged on how Integration of user
profiles 729 much a document satisfied the current information need
or desire that was supposed to be reflected in the query. Obviously this
is a more subjective measure in which pragmatics of documents and
queries play an important role. If the user’s intention is not well
embedded in a query, for example, a retrieved document could be
relevant but not pertinent. Usefulness, finally, was related to the user’s
332

Reference

Disclosure
short-term and/or long-term interests, regardless of the current need
embedded in the query. Thus a pertinent document is expected to be
more or less useful, whereas a useful document may not be pertinent
at all.”
MYAENG, p. 732-34, “Another aspect of the models is examined by
means of user assessments on the general usefulness of retrieved
documents. As summarized in Tables 6 and 7, two prominent rends
are observed across different metric groups (L, is excluded because of
its anomaly,as indicated earlier.) First, almost all profile-based models
appear to be better at retrieving useful documents than M4, regardless
of measures and metrics. This experimental evidence that a profile
alone retrieves more useful documents than a query alone, which is
supposed to represent more direct and short-term information needs,
seems counterintuitive; but it supports the premise that it is difficult to
formulate a query that will reject useless but relevant documents.
Thus, if an IR system is designed to meet a user’s general interests as
well as temporary needs, a query alone does not seem sufficient to
satisfy both demands.
Second, although a profile alone can achieve relatively high
performance in usefulness, it does not necessarily follow that the
existence of query information always reduces satisfaction
(or increases frustration). Instead, it seems essential for models to
include and be guided by some query information in their retrieval
process. As shown in the tables, the models in the Q’ & P category in
the Lz and the inverse cosine metric groups always per-form better
than the profile-alone model when W, is 0.5. In other words, unless the
modified query is very close to the profile, documents retrieved by a
well-balanced retrieval shell are more useful than those retrieved by a
profile or a query alone, or by a shell distorted by emphasis on the
query or profile.”
MYAENG, p. 736-38, “There is little doubt about the importance and
potential advantages of integrating user information into underlying
systems. Especially in information retrieval, the difficulty of
interpreting user queries, which are often incomplete and inaccurate,
necessitates the adaptation of a system to their characteristics. This
research aims at investigating the idea of integrating user interests in
the form of user profile, and establishing a foundation that wili justify
further development in this direction. The analysis of the experimental
results has demonstrated the superiority of profilebased models over a
wide range of criteria and metrics used for evaluation; there were
always some models that outperformed the query alone model.
Although overall effectiveness was improved for those better models,
a dual phenomenon similar to the recall/precision relationship, which
333

Reference

Disclosure
often characterizes information retrieval, also occurred; user
satisfaction evaluated in terms of pertinence appeared to be increased
by integrating a profile, but user frustration was also increased.
However, the integration of user profile improved use fulness in both
satisfaction and frustration. It was particularly noteworthy that for
usefulness almost all profile-based models outperformed the query
alone model. Relevance was used as a device that isolated the
subjective assessments related to the user’s intention from the
objective ones. In spite of the theoretical and intuitive appeal of
Cassini oval over ellipsoidal models, it was difficult to prove the
superiority of the former in general. Instead, Cassini oval models
appeared to be attractive in the L2 metric group, whereas ellipsoidal
models seemed better in the inverse cosine metric group. Although the
results support the main hypothesis and make it possible to select
promising models for more detailed study, the strong regularity in
connection with different parameters and different types of
interactions also suggests further investigation of some aspects of the
model space. There are numerous possible extensions and
improvements to be made in the future. They can be categorized into
three groups: methodologica1 improvements, extensions in
query/profile interactions, and exploration of using profiling tools. In
retrospect, the limitation of resources precluded possibilities of
strengthening the validity of the experimental results; more human
resources could have extended the cutoff point imposed on the number
of documents reviewed by subjects. In addition, by using multiple,
heterogeneous databases and subjects with diverse background, the
query interpretation problem is more likely to surface, and it will be
possible to investigate the roles of user profiles in more realistic and
interesting situations. While there is room for improvement in terms of
more realistic query/profile interaction models, it seems necessary to
connect different user groups with different features of models. This
will make it possible to map different interaction models to different
groups of users and to develop a system that will adapt its query
processing to user characteristics. On the other hand, it would also be
interesting to see relationships between models and the proximity of a
query and a profile in the document space. The third area of research
is concerned with enhancing the quality of user profiles by means of
profiling tools. Two approaches have been explored and are to be
developed further. One is to update user profiles automatically based
on the interaction with users. In this way, more accurate and up-todate user information is expected to be maintained [ 171. Another
approach is based on the finding in psychology that people are better
at recognition than at recall performance [28]. With relationships
among terms available in a given database, the task of formulating a
profile is expected to become less difficult and more effective in that
334

Reference

Disclosure
the task becomes a recognition process rather than a recall [29].”

Sung Myaeng and
Robert Korfhage,
“Towards an Intelligent
and Personalized
Retrieval System”
(“MYAENG II”)

MYAENG II, e.g., Abstract, “Development of an information retrieval
system that can be personalized to each user requires maintaining and
continually updating an interest profile for each individual user. Since
people tend to be poor at self-description, it is suggested that profile
development and maintenance is an area in which machine learning
and knowledge base techniques can be profitably employed. This
paper presents a model for such an application of AI techniques.”
MYAENG II, e.g., 121-22, “In the context of conventional information
retrieval systems (IRS), the search process is initiated and completed
by a set of queries from a user. Each query, usually in the form of a
vector or Boolean expression, consists of a set of key terms to be
matched with the contents of relevant items. To improve the retrieval
effectiveness, modification of the user query through the application
of user feedback has been studied with some successful results [13].
There have also been systems, called selective dissemination of
information systems (SDI), that selectively distribute incoming
information to appropriate users based on user interest profile.
However, only recently has a set of models been proposed that
effectively combines the two different modes of the systems, thereby
attempting to enhance the quality of retrieved items [3,8,9]. One of the
major stumbling blocks in the conventional IRS is the problem of
formulating a query which accurately matches the user's needs and the
contents of potentially relevant items[ 1,12]. Unfortunately, different
users expect different sets of items from the same query and make
different relevance judgements on the same retrieved items, directly
related to their individual needs. But the conventional retrieval system
disregards the individual user's characteristics and the fact that diverse
users have different perceptions of the underlying system. While it is
natural that a user perceives the system in the light of his or her
experience and needs, both the restricted structure of a query and the
nature of the conventional system itself make this perception
unavailable to the system. We believe that knowledge captured in a
user profile embedded in the system will play an essential role in
making a personalized system. One effect can be to retrieve a broader
range of items, some of which would never be brought to the user's
attention on the basis of the query alone. People prefer a librarian who
can surprisingly provide information not explicitly requested but
judged to be important to them. Profile information will also help the
system tailor the retrieved items to a particular user's needs and rank
them appropriately. Again, a friendly and intelligent librarian can
eliminate some information which is not of the user's concern but
would have been retrieved by a novice librarian who had to
335

Reference

Disclosure
relyeffective from the user's point of view and cooperative with the
user in terms of achieving his or her goal. Since we never guarantee
that a user's characteristics and environment stay the same over time,
it becomes necessary for the system to dynamically change the
knowledge kept in the profile. Upon learning various aspects of a
user's information needs and behavior, the system will use this
information to respond in an intelligent and friendly manner. We
elaborate on the concept of a dynamic user profile (DUP) with a
learning strategy for modeling the DUP and discuss the heuristics and
models that utilize the DUP. The next section shows how the system is
configured as a learning system. Our main emphasis in this paper is in
Section 3 where a strategy for learning users' interests and other
characteristics is discussed. The rest of paper, showing the
representation of the DUP, addresses the issues involved in the
utilization of the DUP.”
MYAENG II, e.g., 122-23, “We have developed a full retrieval system
for the purpose of testing the validity and the sensitivity of the
theoretical models with static profiles [8]. This base system can be
modified to reflect the functions of DUP. Since our system should
conduct learning, it is not surprising that its configuration is well
projected on the synthesized model of learning systems proposed by
Smith et al [14]. We adopt terms used in this model to show the
function of each component in the system. The proposed model
consists of six functional components: performance element, instance
selector, critic, learning element, blackboard, and world model. The
performance element uses the learned information to perform the
stated task. The instance selector selects training instances from the
environment of the learning system whereas the critic analyzes the
current abilities of the performance element. The learning element,
which is an essence of the learning system, is an interface between the
critic and the performance element, responsible for translating the
abstract recommendations of the critic into specific changes in rules or
parameters used by the performance element. The blackboard is a
global database used as a system communication means. It holds two
types of informarion: the information in the knowledge base and the
temporary information used by the the learning system components.
Finally, the world model contains the fixed conceptual framework
within which the system operates. Documents in the database are
assumed to contain key words with associated weights. These weights
can be assigned on a frequency-related basis, as is quite standard in
information retrieval. While it is possible to adjust the weights
dynamically on the basis of user response, for present purposes we
assume the weights are fixed. In our system, as shown in the Fig.l, the
query processor/responder is considered as a learning system
336

Reference

Disclosure
performance element. It is the nucleus in conventional systems and,
based on a query, actually retrieves items, providing the user with a
set of items ranked on the basis of the weights in the query and items
in the database. In our system design, this component also integrate
the user-dependent information from the profile. The profile controller
serves mainly as a learning element with some additional functions
taken care of by an instance selector and a critic. This component
observes the interactions between the system and the user, selects
useful instances, and makes specific changes to the profile and
possibly the query in such a way that the system's performance will
eventually approach the desired level. In the context of an IRS, the
role of a critic is performed primarily by human users although the
statistics gathered through operation of the system can be of
importance. Currently, the user's relevance feedback on the retrieved
items is the only valuable information from the critic. Feedback
information from each user is interpreted using the profile, and
therefore part of the critic's role is transferred to the profile
controller.”
MYAENG II, e.g., 123, “Our ultimate purpose in having the learning
element is to build an IRS that incorporates an individual user's
characteristics as much as possible, in an automatic and timedependent manner. Although this can only be achieved by monitoring
the user's interaction with the system, initial dialogue with each user is
expected to play an important role in obtaining skeletal information
that will provide a direction to the system's inference. Without this
kind of information available, the uncertainty we have to deal with is
so high that, either we could never be sure that the system is on the
right track in terms of learning, or the usability of DUP would be
limited. This difficulty will arise especially with users whose
background or interests lie in diverse fields and whose queries are not
consistent with respect to a single field of interest.”
MYAENG II, e.g., 123, “In addition to the need to automatically capture
the user's interest, knowing information regarding individual user's
habits seems also necessary. Typically the following are recognized as
learnable characteristics:
- Reading habits, i.e, preference on the kind of a document (e.g.
theoretical vs. practical)
- Perception on feedback
- Preference on either high recall or high precision
The reading habits can be obtained by simply accumulating statistics.
Given a multidimensional space on which each periodical can be
plotted based on the general trend of its difficulty or practicality, for
example, the learning element of the system extracts the user's
337

Reference

Disclosure
preference along each dimension by observing his feedback on each
document retrieved. If he assesses a document in JACM as pertinent
and/or useful, for instance, the scale about his reading habit should
move toward a more theoretical and difficult document group. The
initial default value can be assigned based on each user's background
information which is expected to be given to the system explicitly.
It is to say that the higher education a user has received, the more
theoretical and difficult documents he would tend to read. On the
other hand, the more he is related to the industry, the more he might
prefer a practical document to theoretical one. A user's perception on
feedback seems to have an implication for any learning strategies.
Since feedback, as a critic, plays an essential role in learning, a user's
general habits on how to assess a retrieved document along different
criteria must be taken into account so that any individual bias can be
eliminated. It is expected that a conservative user will tend to rate a
smaller number of documents positively whereas a more liberal user
will rate a larger number of documents positively. Therefore, the
history on how a user has evaluated retrieved documents will be a
useful source of information. This implication not only facilitates
unbiased learning of a user's characteristics and interest but also
makes it possible to measure system performance more accurately,
taking the bias into account.”

K. Asai, ed. 1995.
Fuzzy Systems for
Information Processing
(1st ed.) (“ASAI”)

ASAI, e.g., at iv, “Fuzzy theory was first used in engineering
applications in the fields of control and information, with many
practical applications and product being brought out, and following
these there was progress in applications in medicine, management,
sociology, natural sciences, psychology and the like. In every case,
approximate models were constructed of the general intelligent
information processing of human beings using fuzzy theory, and using
these either artificial intelligence was created or there were attempts to
explain problems or phenomena that touch mankind. When these
models are constructed, human knowledge, experience, consciousness
and opinions are expressed in natural language, and hearings and
surveys are used. This language is quantified using membership
functions and the information processed, but in most of these cases,
computers are used. Because of this, fuzzy theory has made a
contribution to the extension of conventional computers, which are
based on numbers, to the handling of natural human language.”
ASAI,e.g., at 173, “In the chapters up to now, we have discussed the
current state of hardware and software for fuzzy information
processing. In this chapter, we will discuss fuzzy computers as
systems which extend these and are fuzzy information processing
computers. Fuzzy theory, which is the foundation for these fuzzy
338

Reference

Disclosure
computers, is a means for mathematical quantification of the meaning
of human language, and it is technology indispensable for the
development of human friendly computers.”
Hua Li & Madan Gupta LI, e.g., Table of Contents, Chapter Abstracts, Back Cover, “Fuzzy
(Eds.), Fuzzy Logic and logic offers attractive features for solving many real engineering
Intelligent Systems
problems. As many people have realized, the major obstacles in
(1995) (“LI”)
building a real intelligent machine are dealing with 5 random
disturbances, processing large amounts of imprecise data, interacting
with a dyn~imically changing environment, and 5 coping with
uncertainty. The use of heural-fuzzy techniques can help solve many
of these problems. Fuzzy Logic and Intelligent Systems reflects the
most recent developments in neural networks and fuzzy logic and their
applications in intelligent systems. In addition, the balance between
theoretical work and applications makes this book not only suitable
for researchers and engineers, but for graduate students as well.”

339

Table B6: Fee Records
To the extent the references addressed in claim charts A-1 to A-39 does not disclose the
limitations identified in each chart citing Table B6, one of ordinary skill in the art would be
motivated to combine the references addressed in claim charts A-1 to A-39 with any one or more
of the Table B6 references listed below because: it would have yielded predictable results; using
the techniques of the Table B6 references would have improved the primary or obviousness
references in the same way; and applying the techniques of the Table B6 references to improve
primary or obviousness references would have yielded predictable results.
Reference
U.S. Patent No.
6,119,101
(“PECKOVER”)

Disclosure
See, e.g., PECKOVER, 10:20-29:
A practical and viable electronic marketplace involves the
exchange of market information, as well as the more obvious
trading for goods and services. From a consumer’s point of view,
shopping is a means of gathering data about goods and services
offered. This data is used by the consumer to compare and rank
offerings and to make decisions about purchases. From a provider’s
point of view, consumer shopping is an opportunity to gather data
about consumer needs and interests. This data is used by the
provider to improve product and service offerings.
PECKOVER, 11:16-19:
Consumers have a standardized mechanism for receiving
considerations from advertisers in exchange for allowing delivery
of advertisements and other provider information.
PECKOVER, 11:61-62:
Providers can provide a consideration to consumers for
viewing advertisements and other notices.
PECKOVER, 21:5-11:
A Consideration Account function 67 maintains a “consideration”
account for the user. When the user earns a consideration by, for
example, viewing a directly delivered advertisement or message, or
completing a marketing survey, the consideration amount is
credited to Consideration Account 67. The account is denominated
in a convertible exchange media such as electronic cash tokens.
PECKOVER, 11:44-46:
Advertising may have higher success rates since the targeted
consumers have expressed an interest in the product.
PECKOVER, 11:54-64:
The mechanism for quantifying consumer demand uses data based
340

Reference

U.S. Patent No.
5,105,184
(“PIRANI”)

Disclosure
on individual buying decisions, not merely aggregate or estimated
data.
Providers can quantify demand in real-time.
Providers have a mechanism for discovering the reasons for lost
sales.
Providers can provide a consideration to consumers for viewing
advertisements and other notices.
Providers can receive feedback in real-time about the success of
promotions.
PECKOVER, 20:13-19:
A Decision Agent Archive 80 stores and accesses Decision Agents
14 that are expired, i.e., agents that have completed their tasks,
whether successfully or not. For example, if a Demand Agent 16
needs more detailed data about a query than is stored in a Query
Logger 136 of a Market 18, it can access the details of the related
Decision Agent 14 through Decision Agent Archive 80.
PECKOVER, 18:40-53:
Referring to FIG. 4A, a Personal Agent 12 or 13 comprises the
functional components of:
a Unique identification (ID) 50,
an Owner Manager 52,
a Preference Manager 54,
a Delivery Manager 56,
an Individual Firewall 58,
a Decision Agent Manager 60,
a Demand Agent Manager 62,
an Ad Manager 64,
a Target Manager 66, and
a Consideration Account 67.
PECKOVER, 29:49-67:
The Decision Agent’s Response Manager 108 collects references
(step 326) to the matching ads found by Basic Search Engine. The
Response Manager also sends a response to the Personal Agent that
placed the advertisement (if the placer so desired and marked in the
ad), providing real-time feedback to the placer. Immediate Agents
then removes the Decision Agent from its internal queue and gives
the Decision Agent back to Active Decision Agent Manager 152
(step 328).
PIRANI, 3:1-7:
This new use can also provide to a small or a new software
developer much needed help to launch a software project. By
convincing the viability of the project to a commercial company
which advertise widely to sell their products, the software
developer can receive revenue from such company in exchange for
the right to advertise in the new software.
341

Reference
U.S. Patent No.
5,710,884
(“DEDRICK
PATENT”)

Disclosure
DEDRICK PATENT, 10:8-21:
Thus, the metering server 14 contains an account balance, a user
identification (such as an account number or a name), and may also
include information indicating which information the user
subscribes to. User profile data requested by metering server 14
from the client systems 12 is stored in user profile database 30,
along with user profile data corresponding to electronic information
being consumed by an end user. As discussed above, this user
profile data does not specifically identify the individual end user.
The account balance and user identification is contained in the
transaction database 32. Therefore, the only information which is
contained in the metering server which identifies an individual end
user is that user’s identification and an account balance, thereby
maintaining the user’s privacy.
DEDRICK PATENT, 10:22-29:
In one embodiment, the transaction database 32 also includes, in the
log of a transaction, an indicator of the electronic information
consumed. By maintaining such a log, the metering server 14 is
able to summarize an end user’s consumption for that user’s
review. For example, the metering server 14 may generate a
monthly statement summarizing how much money the end user
spent consuming electronic information.
DEDRICK PATENT, 10:45-61:
If the end user is not a subscriber, the metering process 36
calculates the price of the requested information and accesses the
transaction database to subtract the price from the balance of the
end user’s account. The balance is initially established when the
end user requests an account in the system. The balance may be
specified by the end user and approved by the clearinghouse server.
Approval may be based upon a credit card nmnber or bank account
number provided by the end user. The balance may be updated by
the clearinghouse server when the end user pays his bill. If the
balance minus price is greater than zero, the metering process 36
retrieves the information and sends the same to the end user. If the
balance minus price is less than zero, the metering process 36 does
not retrieve the information and may send a message to the end user
that the balance has been exceeded. The initial balance of the
account is typically set by a credit limit.
DEDRICK PATENT, 11:35-55:
The software tools include “cost type” and “cost value” fields that
accompany each unit of electronic information. The cost type and
cost value can be utilized to calculate a price that can be either
credited to or debited from the end users. The fields allow the
publisher/advertiser 18 to establish the manner in which the
information will be charged to the end user’s account. One example
342

Reference

Disclosure
of a cost type is “pay per view” payment method, wherein the end
user pays an associated cost each time the user consumes a unit of
information. This cost may also be proportional to the amount
consumed, so that the cost is higher for consuming the entire unit
infonnation rather than a smallm portion, such as the abstract. This
type of payment may be desirable for information which is typically
seldom consumed by the end user. Other cost types include
payment on a per byte or word of information viewed by the end
user, or payment for the period of time that the user consumes the
information. These cost types may be desirable when the end user is
accessing a database that contains, for example, corporate or
individual credit information, or the drawings and text of a patent
database.
DEDRICK PATENT, 12:1-26:

DEDRICK PATENT, 12:43-54:
The publisher/advertiser is also provided with an account number
so that the charges associated with the consumption of information
provided by the publisher/advertiser is charged to the account
number of the publisher/advertiser. For example, a publisher may
provide a unit of information which is subsequently consumed by
the end user. The charge incurred by the end user is then debited
against the user’s account and credited to the publisher’s account.
By way of another example, the end user may view an
advertisement, wherein the charge associated with the unit of
information viewed is credited to the end user’s account and
debited to the advertiser’s account.
DEDRICK PATENT, 14:19-37:
As shown in FIG. 4, each clearinghouse server 20 contains a
343

Reference

Disclosure
demographic database 50, a transaction database 52, billing process
54 and a session manager 56. The demographic database 50
contains user profile data collected from the metering servers 14.
The transaction database 52 contains billing information relating to
the end users. The transaction database 52 also contains data
relating to the accounts of the publishers/advertisers 18. The billing
process 54 can access and process data within the databases 50 and
52. For example, when an end user consumes a unit of electronic
information, data relating to the consumption of the electronic
information may be sent from the billing server 14 to the
clearinghouse server 20. The session manager 56 instructs the
billing process 54 to charge the publisher/advertiser account within
the transaction database 52. The clearinghouse server 20 may also
receive user profile data from the metering servers 14 which is
subsequently stored by the billing process 54 in the demographic
database 50.
DEDRICK PATENT, 15:7-25:
In one embodiment, the billing process 54 also generates bills for
the end users and the publishers/advertisers. Upon a request from
the publisher/advertiser, the session manager 56 instructs the billing
process 54 to generate a bill. The billing process 54 retrieves the
billing information from the transaction database 52 and generates a
bill. The bill may be electronically transferred to the end user or
sent through a conventional mail service. The billing process 54
may also generate bills that are transmitted to the publishers
advertisers. The bill may be generated periodically in accordance
with header information that accompanies the content that is
generated by a publisher/advertiser. Alternatively, the
clearinghouse server 20 may utilize consumer credit cards and or
bank accounts for billing. For example, amounts owed by the
consumer for consumption of electronic content and amounts due
the consumer for consumption of electronic advertisements may be
charged or credited, respectively, to the consumer's credit card or
bank account.
DEDRICK PATENT, 17:13-26:
The metering server 14 in conjunction with the client activity
monitor 24 of the client system may monitor the end user’s
consumption of electronic advertising information and provide user
profile data to the metering server 14 relating to the end user. For
example, the metering process 36 may monitor the amount of time
an end user spends viewing an electronic advertisement, or which
particular advertisement or page of the advertisement was of
interest to the end user. The metering process 36 may further
monitor what answers were provided by the user, or paths taken by
the user in an interactive model, along with follow-up requests
344

Reference

U.S. Patent No.
7,072,849
(“FILEPP”)

Disclosure
initiated by the end user in an interactive model. This information is
then forwarded to the clearinghouse server 20 for compilation.
See, e.g., FILEPP, 3:1-4:
And, it is still a further object of this invention to provide a method
for presenting advertising in an interactive service which method
enables the user to transactionally interact with the advertising
presented.
FILEPP, 3:44-67:
Also in preferred form, the method includes step for maintaining an
advertising object identification queue, and an advertising object
store that are replenished based on predetermined criteria as
advertising is called for association and presentation with
applications. In accordance with the method, as applications are
executed at the reception system, the application objects provide
generalized calls for advertising. The application calls for
advertising are subsequently forwarded to the reception system
advertising queue management facility which, in turn supplies an
identification of advertising who’s selection has been
individualized to the user based on, as noted, the user’s prior
interaction history with the service, demographics and local.
Thereafter, the object identification for the advertising is passed to
the object store to determine if the object is available at the
reception system. In preferred fonn, ifthe advertising object is not
available at the reception system, a sequence of alternative
advertising object identifications can be provided which if also are
unavailable at the reception system will resulting in an advertising
object being requested from the network. In this way, advertising of
interest can be targeted to the user and secured in time-efficient
manner to increase the likelihood of user interest and avoid service
distraction.
FILEPP, 7:27-32:
In preferred form, network 10 provides information, advertising and
transaction processing services for a large number of users
simultaneously accessing the network via the public switched
telephone network (PSTN), broadcast, and/or other media with their
RS 400 units. Services available to the user include display of
information such as movie reviews, the latest news, airlines
reservations, the purchase of items such as retail merchandise and
groceries, and quotes and buy/sell orders for stocks and bonds.
Network 10 provides an environment in which a user, via RS 400
establishes a session with the network and accesses a large number
of services. These services are specifically constructed applications
which as noted are partitioned so they may be distributed without
undue transmission time, and may be processed and selectively
stored on a user’s RS 400 unit.
345

Reference
FLYNN

U.S. Patent Nos.
5,948,061
(“MERRIMAN I”)
and 7,844,488
(“MERRIMAN II”)

Disclosure
See e.g., FLYNN, p. 1 (“Once they begin running ads on various sites,
advertisers analyze the number of times somebody clicked on their ad, then
change the placement or timing of their ad to try and improve the ‘click
rate.’”)
See, e.g., MERRIMAN I (AND CORRESPONDING DISCLOSURE IN MERRIMAN
II), 2:59-3:4:
The basic architecture of the network 10 comprises at least one
affiliate web site 12, an advertisement (ad) server web site 19 and
one or more individual advertiser’s web sites 18. Affiliates are one
or more entities that generally for a fee contract with the entity
providing the advertisement server permit third party
advertisements to be displayed on their web sites. When a user
using a browser accesses or “visits” a web site of an affiliate, an
advertisement provided by the advertisement server 19 will be
superimposed on the display of the affiliate’s web page displayed
by the user’s browser. Examples of appropriate affiliates include
locator services, service providers, and entities that have popular
web sites such as museums, movie studios, etc.
MERRIMAN I (AND CORRESPONDING DISCLOSURE IN MERRIMAN II), 3:5-23:
The basic operation of the system is as follows in the preferred
embodiment. When a user browsing on the Internet accesses an
affiliate’s web site 12, the user’s browser generates an HTTP
message 20 to get the information for the desired web page. The
affiliate’s web site in response to the message 20 transmits one or
more messages back 22 containing the information to be displayed
by the user’s browser. In addition, an advertising server process 19
will provide additional information comprising one or more objects
such as banner advertisements to be displayed with the information
provided from the affiliate web site. Normally, the computers
supporting the browser, the affiliate web site and the advertising
server process will be at entirely different nodes on the Internet.
Upon clicking through or otherwise selecting the advertisement
object, which may be an image such as an advertisement banner, an
icon, or a video or an audio clip, the browser ends up being
connected to the advertiser’s server or web site 18 for that
advertisement object.
MERRIMAN I (AND CORRESPONDING DISCLOSURE IN MERRIMAN II), 3:24-63:
In FIG. 1, a user operates a web browser, such as Netscape or
Microsoft Internet Explorer, on a computer or PDA or other
Internet capable device 16 to generate through the hypertext
transfer protocol (HTTP) 14 a request 20 to any one of preferably a
plurality of affiliate web sites 12. The affiliate web site sends one or
more messages back 22 using the same protocol. Those messages
22 preferably contain all of the information available at the
particular web site 12 for the requested page to be displayed by the
346

Reference

Disclosure
user’s browser 16 except for one or more advertising objects such
as banner advertisements. These objects preferably do not reside on
the affiliate’s web server. Instead, the affiliate’s web server sends
back a link including an IP address for a node running an advertiser
server process 19 as well as information about the page on which
the advertisement will be displayed. The link by way of example
may be a hypertext markup language (HTML) <img> tag, referring
to, for example, an inline image such as a banner. The user’s
browser 16 then transmits a message 23 using the received IP
address to access such an object indicated by the HTML tag from
the advertisement server 19. Included in each message 23 typically
to the advertising server 19 are: the user’s IP address, (ii) a cookie
if the browser 16 is cookie enabled and stores cookie information,
(iii) a substring key indicating the page in which the advertisement
to be provided from the server is to be embedded, and (iv) MIME
header information indicating the browser type and version, the
operating system of the computer on which the browser is operating
and the proxy server type. Upon receiving the request in the
message 23, the advertising server process 19 determines which
advertisement or other object to provide to user’s browser and
transmits the messages 24 containing the object such as a banner
advertisement to the user’s browser 16 using the HTTP protocol.
Preferably contained within the HTTP message is a unique
identifier for the advertiser’s web page appropriate for the
advertisement. That advertisement object is then displayed on the
image created by the web user’s browser as a composite of the
received affiliate’s web page plus the object transmitted back by the
advertising web server.
MERRIMAN I (AND CORRESPONDING DISCLOSURE IN MERRIMAN II), Fig. 1:

347

Reference

ADSERVER 2.0

ADSERVER 2.0; AD
REPORTING

NETGRAVITY
ADSERVER HELP

ABOUT
NETGRAVITY
ADSERVER

Disclosure

MERRIMAN II (AND CORRESPONDING DISCLOSURE IN MERRIMAN II), 9:3841:
2. The method of claim 1, wherein selecting an advertisement based
upon stored information about said user node comprises selecting
an advertisement based upon a prior content request sent from said
user node to an affiliate node.
See e.g., ADSERVER 2.0, p. 2 (“By tracking viewer response to advertising,
NetGravity reports allow agencies and advertisers to quickly test the
effectiveness of their campaigns.”)
See e.g., ADSERVER 2.0; AD REPORTING, p. 1 (“Performance is indicated by
the number of ad impressions and click-throughs for ads and advertisers.”);
id. (describing that reports are provided on impressions/clicks.); id.
(“AdServer supports premium ad types, the ability to test different ads in
real-time, and the delivery of reliable performance reports.”); id., p. 2 (“By
tracking viewer response to advertising, NetGravity reports allow agencies
and advertisers to quickly test the effectiveness of their campaigns. Such
rapid and reliable feedback empowers advertisers with the information they
need to maximize their advertising efforts.”)
See e.g., NETGRAVITY ADSERVER HELP, Installing the Redirection Utility
(“When a visitor to your site clicks on an ad, AdServer redirects them to
the advertiser’s site. Before they go there, however, AdServer must record
that they clicked on the ad.”); see also id., AdSpace Specs, Working with
Ads; id., AdServer Tools Reference (“RepAd – generates ad reports.”)
See e.g., ABOUT NETGRAVITY ADSERVER, Getting Started, p. 1 (“AdServer
records when the ad is shown, and also when it is clicked. You can then
generate reports that show ad and location performance.”); id., p. 3:
348

Reference

NETGRAVITY
ADSERVER
CHOSEN BY GNN

Disclosure
“Instead of immediately sending a user to the advertiser’s site, all ad links
automatically execute the redir program. This is a CGI program that first
records the user’s click before redirecting the user’s browser to the
advertiser’s site.”); id., Serving Ads Dynamically, p. 2 (“. . . 8. The visitor
views the page and the ad. When they click on the ad, they issue a call to
the redirect utility on your content server. 9. The redirect utility records
the user’s click in the AdServer logs, then sends the user to the advertiser’s
site.”); id., Serving Ads Dynamically, p. 5 (“When AdServer serves an ad,
it records in the AdServer_log file that the ad has been shown. Similarly,
the redirect utility records that an ad was clicked by writing to the
AdServer_log. . . . During its normal operation, AdServer writes to the
AdServer_log file each time an ad is requested, and each time the redirect
utility is notified that an ad has been clicked.”); id., Serving Ads
Dynamically, p. 6 (“The parselog tool reads the AdServer_log file, extracts
statistics about which ads received impressions and clicks, and writes that
information to the AdServer database.”); id., AdServer Tools, p. 2:
“Parselog reads your content server’s log file and writes usage statistics
into the AdServer database. AdServer uses this information to measure the
number of impressions and clicks an ad has received.” ); see also id., p. 5
(same), p. 6 (same).); id., Internal Specifications, p. 9 (listing logging “the
number of clicks received”), p. 11 (listing that the system records that a
“dynamically served ad received an impression” and that a “dynamically
served ad received a click”); id., NGAPI Function Reference, p. 22 (noting
that the ID of the ad that is clicked is logged), p. 23 (“records that an ad
was clicked”), p. 37 (records “the number of clicks received”), p. 42
(same)
See e.g., NETGRAVITY ADSERVER CHOSEN:
NetGravity, the leader in Internet advertising technology, today
announced GNN, a service of America Online Inc., will take
advantage of the NetGravity AdServer technology for WebCrawler,
its Internet search service. This allows GNN to better manage its
WebCrawler advertising inventory, dynamically deliver targeted
ads, measure advertising results in real time, and automate ad sales
efforts. As part of this agreement, GNN becomes the first company
to capitalize on the alliance between NetGravity and I/Pro (Internet
Profiles Corporation), the leading Internet measurement firm. This
builds on GNN's longstanding relationship with I/Pro and enhances
its ability to provide the most comprehensive reports on advertising
effectiveness and to deliver them to advertisers far faster than sites
not using the NetGravity technology.
NetGravity was founded to enable Web publishers to retain
complete control of their online advertising management. Unlike
other companies which merely provide services for ad placement
and scheduling, NetGravity offers a unique approach, providing the
software and technology which empowers publishers to manage
349

Reference

Disclosure
advertising inventory, dynamically target ads to the right audiences,
measure results in real time, and automate sales efforts. Now,
through NetGravity's relationship with I/Pro, Web sites will be able
to develop and place advertising much more effectively using
management tools with demographic profiles for targeted ad
placement. Sites using the NetGravity AdServer are able to retain
all advertising revenues and eliminate the risks of dependency on
external services such as downtime, increasing costs, unplanned
maintenance and unpredictable management.

“For More About
Tide, Click Here”
by Zachary
Schiller,
Bloomberg
Businessweek, June
2, 1996.
(“SCHILLER”)

See e.g., SCHILLER: “In a test arrangement, instead of compensating online
companies for each consumer who sees a P&G ad, P&G will pay only
when the online customer ‘clicks’ from that ad to one of P&G’s own Web
sites. This means that Yahoo!, a major online provider that agreed to
P&G’s terms, won’t make any money if a customer sees a spot promoting
P&G’s SunnyDelight juice drink unless the customer moves on to its
Sunny Delight Web site, which has a game with various prizes.”

DEDRICK 1994

See e.g., DEDRICK 1994, p. 57 (“Soon however, advertisers will be more
attracted to a distribution medium that . . . provides proof back to the
advertiser showing aggregate consumption statistics for an advertisement”
); id. (p. 57: “The advertisers will pay for the storage and distribution
services of the yellow pages, based upon the quality of the targeted
consumers currently served by the yellow pages.”);id., p. 59 (“Paying for
usage of the electronic yellow pages may follow a variety of models. One
likely model consists of the advertiser paying the electronic yellow pages
service provider a fee for storing and distributing each advertisement for a
specified period of time.”); id., p. 61 (“Electronic content metering
capabilities must exist within the servers that communicate with the client
consumption devices. This will enable charging consumers for electronic
content consumption and to pay the same consumer rebates for the
consumption of electronic advertisements. . . . Some metering
methodologies that may be important are pay per view of object (same cost
each time or a decreasing cost based upon number of views), pay per byte
(or other designated unit of content),pay per second (or other designated
unit of time) . . .”); id., p. 62: “Specifically, the currently suggested
attribute extension list is as follows: . . . Metering methodology attributes
(includes debit and credit capabilities), Metering methodology pricing
attributes”)
See e.g., DEDRICK 1995, p. 42 (“provides statistics to advertisers showing
aggregate consumption for an advertisement.”); id. (“Advertisers will pay
for storage and distribution services based on the quality of the targeted
consumers currently served by the yellow pages.”); id., p. 44 ("Paying for
use of the electronic Yellow Pages could follow a variety of models. One

DEDRICK 1995

350

Reference

GALLAGHER

NETGRAVITY
ADSERVER
CHOSEN BY GNN

Lycos, Inc.
Registration
Statement No. 333354, dated April 3,
1996 (“LYCOS
PROSPECUS”),
produced at
GOOG-WRD00872476-GOOGWRD-00872549

Disclosure
likely model consists of the advertiser paying the electronic Yellow Pages
service provider a fee for storing and distributing each ad for a specified
period of time." )
See e.g., GALLAGHER, p. 7 (“Profiles accommodate the possibility that
some users within the region of acceptability may be more desirable to an
advertiser than others. Hen, a distance metric capturing the relative
desirability of a user with respect to an ideal profile is possible. . . .
recognizing a notion of distance allows the possibility for advertisers to
‘bid’ for the opportunity to display an advertisement to a user. Such bids
would be determined by the advertiser, based on variables such as the user
profile . . . and advertising budget.”); id., p. 8 (“When bids are received,
they can be ranked. The banner advertisement corresponding to the
winning bid is displayed to the user. Other advertisements may be
displayed according to their ranking if there is an opportunity to display
additional advertisements (e.g., if the user engages in several search or
browse activities during a session).”)
NETGRAVITY ADSERVER CHOSEN BY GNN (NetGravity, the leader in
Internet advertising technology, today announced GNN, a service of
America Online Inc., will take advantage of the NetGravity AdServer
technology for WebCrawler. . . . This allows GNN to . . . measure
advertising results in real time . . .”)
See LYCOS PROSPECTUS at GOOG-WRD-00872492:

Id. at GOOG-WRD-00872503-504:

351

Reference

Disclosure

Id. at GOOG-WRD-00872503-505:

Id. at GOOG-WRD-00872535:

Lycos, Inc. Form
S-1 Registration
Statement, dated
February 14, 1996
(“LYCOS S-1”),
produced at
GOOG-WRD00872550-GOOGWRD-00872923

See LYCOS S-1 at GOOG-WRD-00872568:

Id. at GOOG-WRD-00872579:

352

Reference

Disclosure

Id. at GOOG-WRD-00872580:

Id. at GOOG-WRD-00872581:

Id. at GOOG-WRD-00872609:

353

Reference
Disclosure
Excite, Inc. SB-2
Registration
Statement No. 3332328-LA, March
11, 1996 (“Excite
SB-2”) produced at
GOOG-WRD00872006-GOOGId. at GOOG-WRD-00872037.
WRD-00872094

Id. at GOOG-WRD-00872039.

Id. at GOOG-WRD-00872043.

Id.

354

Reference

Disclosure

Id. at GOOG-WRD-00872044.

Id.
Excite, Inc.
Prospectus, dated
April 3, 1996
(“Excite
Prospectus”)
produced at
GOOG-WRD00871928GOOGL-WRD00872005

Id. at GOOG-WRD-00871957.

Id. at GOOG-WRD-00871959.

355

Reference

Disclosure

Id. at GOOG-WRD-00871963.

Id.

Id. at GOOG-WRD-00871964.

Id.
356

Reference

Disclosure

InfoSeek
Corporation S-1
Registration
Statement No. 3334142, Amendment
No. 1, dated May
3, 1996 (“InfoSeek
Id. at GOOG-WRD-00872378.
S-1”) produced at
GOOG-WRD00872371-GOOGWRD-00872464

Id. at GOOG-WRD-00872380.

Id. at GOOG-WRD-00872388.

Id. at GOOG-WRD-00872402-403.

357

Reference

Disclosure

Id. at GOOG-WRD-00872404.

Id. at GOOG-WRD-00872404-05.

Id. at GOOG-WRD-00872405.

Id. at GOOG-WRD-00872409-10.

358

Reference

Disclosure

Id. at GOOG-WRD-00872410.

Id. at GOOG-WRD-00872410-411.

Id. at GOOG-WRD-00872411.

359

Reference
Disclosure
Yahoo Prospectus
Registration
Statement No. 3332142, dated April
12, 1996 (“Yahoo
Prospectus”)
produced at
GOOG-WRD00874251-GOOGWRD-00874328

Id. at GOOG-WRD-00874253.

360

Reference

Disclosure

GOOG-WRD-00874258-59.

Id. at GOOG-WRD-00874261.

Id. at GOOG-WRD-00874264-65.

361

Reference

Disclosure
Id. at GOOG-WRD-00874262.

Id. at GOOG-WRD-00874275.

Id. at GOOG-WRD-00874280.

Id. at GOOG-WRD-00874289.

362

Reference

Disclosure

Id. at GOOG-WRD-00874316.
Yahoo Form SB-2
Registration
Statement No. 3332142, dated March
7, 1996 (“Yahoo
Form SB-2”)
produced at
GOOG-WRD00874329-GOOGWRD-00874418

Yahoo Form SB-2 at GOOG-WRD-00874333.

363

Reference

Disclosure

GOOG-WRD-00874337-38.

Id. at GOOG-WRD-00874339-40.

364

Reference

Disclosure

Id. at GOOG-WRD-00874343.

Id. at GOOG-WRD-00874369.

Id. at GOOG-WRD-00874353.

365

Reference

Disclosure

Id. at GOOG-WRD-00874358.

Id. at GOOG-WRD-00874366-67.

Id. at GOOG-WRD-00874393.

366

Reference
Disclosure
Open Text Form F1 Registration
Statement No. 3398858, dated
November 1, 1995
(“Open Text Form
F-1”) produced at
GOOG-WRD00873727-GOOGWRD-00873878

Id. at GOOG-WRD-00873739.

Id. at GOOG-WRD-00873775.

Id. at GOOG-WRD-00873779.

367

Reference

Disclosure
Id. at GOOG-WRD-00873783-84.

Open Prospectus,
dated January 23,
1996 (“Open Text
Prospectus”)
produced at
OT03652-3758

Id. at OT03658.

Id. at OT03694.

Id. at OT03698.

368

Reference

Disclosure
Id. at OT037002-03.

369

Table B7: Databases, Clients, Servers
To the extent the references addressed in claim charts A-1 to A-39 does not disclose the
limitations identified in each chart citing Table B7, one of ordinary skill in the art would be
motivated to combine the references addressed in claim charts A-1 to A-39 with any one or more
of the Table B7 references listed below because: it would have yielded predictable results; using
the techniques of the Table B7 references would have improved the primary or obviousness
references in the same way; and applying the techniques of the Table B7 references to improve
primary or obviousness references would have yielded predictable results.
Reference
PECKOVER

Disclosure
See, e.g., PECKOVER, 17:6-10:
Various specialized agents are described in conjunction with
other Figures. Agents and other components operating in
Agent Marketplace 28 have access to a Product Database
(Product DB or PDB) 32.
PECKOVER, 23:17-20:
A Product Listing function 124 maintains a list of the products
that can be advertised in this market. Each product references
detailed product data that is kept in a Product Database (PDB)
32 described in conjunction with FIG. 9A.
PECKOVER, 23:43-47:61:
An Active Ads function 146 maintains the ads that are
currently active. As each new add is accepted by Active Ads
function 146, an Active Decision Agent Manager 152 (see
below) is notified so that pending searches can be matched
against the new advertisement.
PECKOVER, 25:10-36:
A Remote Database Adaptor 140 provides communication and
session management services to connect to a database (a
“remote database”, not shown) belonging to a manufacturer or
a provider. Remote Database Adaptor 140 also provides
translation services to translate between the data formats used
by a remote database and the data formats used by PDB 32.
Remote Database Adaptor 140 allows a provider to submit ads
directly from the provider’s remote database into Market 18.
Remote Database Adaptor 140 also allows access “by
reference” to advertisement data that remains stored in a
remote database; that is, the data is not copied into Agent
System 10, but is accessed as needed. Market 18 includes a
370

Reference

U.S. PATENT NO.
5,710,884 (“DEDRICK
PATENT”)

Disclosure
Remote Database Adaptor 140 for each provider that chooses
to supply ads in this manner; alternatively, a provider uses
various functional components accessed via provider’s
Personal Agent 13 to place ads manually.
PECKOVER, 25:36-57:
Referring to FIG. 9A, a Product Database 32 (PDB) comprises
functional components:
a Database Administration function 166,
a Product Data Storage function 168,
a Product Template Manager function 170,
and, (optionally) some number of Remote Database
Adaptors 172.
PDB 32 maintains generic data about products, to be
referenced by ads placed by providers. Although PDB 32 is
illustrated here as a single database (with several internal
components) for ease of understanding, the contemplated PDB
32 will be split across several processors 38, as illustrated
previously in FIG. 3A.
Referring to FIG. 9A, a Database Administration function 166
provides conventional add, delete, update, query, and backup
access for a System Administrator user to the other
components of PDB 32.
A Product Data Storage function 168 stores data about
different products, for example, product name, product model
number, manufacturer’s suggested retail price for product, etc.
See, e.g., DEDRICK PATENT, 3:37-44:
The metering server 14 is coupled to a publisher unit 18
through a plurality of clearinghouse servers 20. By way of
example, the publisher 18 may be connected to the server 14 as
part of an overall wide area network (WAN) that allows the
server 14 and publisher unit 18 to transfer information. The
system 10 may also have a yellow page server 22 coupled to
the publisher unit 18 and the metering servers 14. The
publisher unit and servers of the WAN system contain the
interface hardware and software necessary to transfer
electronic information between the components of the system.
As shown in FIG. 1, the system 10 may have multiple client
systems 12 coupled to a single metering server 14 and multiple
servers 14 coupled to a single clearinghouse server 20, a
regional content database server 21 and a single yellow page
server 22. There may be multiple clearinghouse and yellow
page servers located at regional centers throughout the
country/world. In addition, depending on the size of a
community, there may also be multiple yellow page servers for
each local community. Although the computer 18 is referred to
371

Reference

U.S. Patent No.
6,374,237 (“REESE”)

Disclosure
as a publishing unit, it is to be understood that the computer
can also be a node for an advertiser 18 and that the use of the
terms publisher and advertiser may be synonymous.
DEDRICK PATENT, 5:39-51:
Session manager 29 transfers data and control information to
and from the components of client system 12, and acts as an
interface between client system 12 and metering server 14.
Electronic information which is transferred to client system 12
is received by session manager 29 and forwarded to client
interface 23. In one embodiment, the electronic information. is
forwarded to client interface 23 via content adapter 25. Content
adapter 25 may then modify the electronic information, based
on the end user’s data stored in personal profile database 27.
Session manager 29 also instructs statistic compilation process
26 to compile the aggregate data stored in personal profile
database 27 when the information is requested by metering
server 14.
DEDRICK PATENT, 7:28-39:
Data is collected for personal profile database 27 by direct
input from the end user and also by client activity monitor 24
monitoring the end user’s activity. When the end user
consumes a piece of electronic information, each variable (or a
portion of each variable) within the header block for that piece
of electronic information is added to the database for this end
user. For example, if this piece of electronic information is
made available to the end user for consumption in both audio
and video format, and the end user selects the audio format,
then this choice of format selection is stored in personal profile
database 27 for this end user.
REESE, 1:12-21:
The World Wide Web brings the vast amount of information
on the Internet to the public's attention. A problem today in
web browsing is that browsing is essentially flat, with no
semantic meaning applied to query and search mechanisms.
Between the client, an application program that establishes
connections for the purpose of sending requests from a user,
and the server, an application that accepts connections in order
to service requests by sending back responses, there exists a
bandwidth problem of not being able to get information
quickly enough to the user on the client end to do meaningful
operations.
REESE, 2:49-65:
FIG. 1 presents a block diagram of the invention. FIG. 1 shows
a client 110 that is an application program that establishes
connections for the purpose of sending requests to a matching
372

Reference

U.S. Patent Nos.
5,948,061 (“MERRIMAN
I”) and 7,844,488
(“MERRIMAN II”)

Disclosure
server 120. The client 110 contains a user agent that initiates
the request. The user agent is, for example, a browser, editor,
spider (web-traversing robot), or other end user tool that can
service different requests by a user. Typical browsers include
NETSCAPE NAVIGATORTM or INTERNET EXPLORERTM.
The matching server 120 is an application program that accepts
connections in order to service requests by sending back
responses. In the case of a browser, a request is sent in a
typical protocol, for example, hypertext transfer protocol
(HTTP). Other protocols include Simple Mail Transfer
Protocol (“SMTP”), Network News Transfer Protocol
(“NNTP”), File Transfer Protocol (“FTP”), Gopher, and Wide
Area Information Service (“WAIS”).
REESE, 6:54-67:
FIG. 8 presents a flow chart of the construction of a matching
server database of the invention. In FIG. 8, a matching server
is designated. In step 800, a matching server is designated to
construct an aggregate database. In step 810, a list of content
servers is designated from which to collect data that will make
up the aggregate data of the matching server. The content
servers designated could be any or all servers in an Internet
environment or select servers in an Intranet or other network
environment. Next, in step 820, the matching server walks
each of the content servers and collects information that will
make up the aggregate database. Next, in step 830, the
matching server builds an aggregate database that is a
representation of the content servers walked.
See, e.g., MERRIMAN I (AND CORRESPONDING DISCLOSURE IN
MERRIMAN II), 2:59-3:4:
The basic architecture of the network 10 comprises at least one
affiliate web site 12, an advertisement (ad) server web site 19
and one or more individual advertiser’s web sites 18. Affiliates
are one or more entities that generally for a fee contract with
the entity providing the advertisement server permit third party
advertisements to be displayed on their web sites. When a user
using a browser accesses or “visits” a web site of an affiliate,
an advertisement provided by the advertisement server 19 will
be superimposed on the display of the affiliate’s web page
displayed by the user’s browser. Examples of appropriate
affiliates include locator services, service providers, and
entities that have popular web sites such as museums, movie
studios, etc.
MERRIMAN I (AND CORRESPONDING DISCLOSURE IN MERRIMAN II), 3:523:
The basic operation of the system is as follows in the preferred
373

Reference

Disclosure
embodiment. When a user browsing on the Internet accesses
an affiliate’s web site 12, the user’s browser generates an
HTTP message 20 to get the information for the desired web
page. The affiliate’s web site in response to the message 20
transmits one or more messages back 22 containing the
information to be displayed by the user’s browser. In addition,
an advertising server process 19 will provide additional
information comprising one or more objects such as banner
advertisements to be displayed with the information provided
from the affiliate web site. Normally, the computers supporting
the browser, the affiliate web site and the advertising server
process will be at entirely different nodes on the Internet. Upon
clicking through or otherwise selecting the advertisement
object, which may be an image such as an advertisement
banner, an icon, or a video or an audio clip, the browser ends
up being connected to the advertiser’s server or web site 18 for
that advertisement object.
MERRIMAN I (AND CORRESPONDING DISCLOSURE IN MERRIMAN II),
3:24-63:
In FIG. 1, a user operates a web browser, such as Netscape or
Microsoft Internet Explorer, on a computer or PDA or other
Internet capable device 16 to generate through the hypertext
transfer protocol (HTTP) 14 a request 20 to any one of
preferably a plurality of affiliate web sites 12. The affiliate
web site sends one or more messages back 22 using the same
protocol. Those messages 22 preferably contain all of the
information available at the particular web site 12 for the
requested page to be displayed by the user’s browser 16 except
for one or more advertising objects such as banner
advertisements. These objects preferably do not reside on the
affiliate’s web server. Instead, the affiliate’s web server sends
back a link including an IP address for a node running an
advertiser server process 19 as well as information about the
page on which the advertisement will be displayed. The link by
way of example may be a hypertext markup language (HTML)
<img> tag, referring to, for example, an inline image such as a
banner. The user’s browser 16 then transmits a message 23
using the received IP address to access such an object
indicated by the HTML tag from the advertisement server 19.
Included in each message 23 typically to the advertising server
19 are: the user’s IP address, (ii) a cookie if the browser 16 is
cookie enabled and stores cookie information, (iii) a substring
key indicating the page in which the advertisement to be
provided from the server is to be embedded, and (iv) MIME
header information indicating the browser type and version,
374

Reference

U.S. Patent No.
7,072,849 (“FILEPP”)

Disclosure
the operating system of the computer on which the browser is
operating and the proxy server type. Upon receiving the
request in the message 23, the advertising server process 19
determines which advertisement or other object to provide to
user’s browser and transmits the messages 24 containing the
object such as a banner advertisement to the user’s browser 16
using the HTTP protocol. Preferably contained within the
HTTP message is a unique identifier for the advertiser’s web
page appropriate for the advertisement. That advertisement
object is then displayed on the image created by the web user’s
browser as a composite of the received affiliate’s web page
plus the object transmitted back by the advertising web server.
MERRIMAN I (AND CORRESPONDING DISCLOSURE IN MERRIMAN II), Fig.
1:

MERRIMAN II (AND CORRESPONDING DISCLOSURE IN MERRIMAN II),
9:38-41:
2. The method of claim 1, wherein selecting an advertisement
based upon stored information about said user node comprises
selecting an advertisement based upon a prior content request
sent from said user node to an affiliate node.
See, e.g., FILEPP, 5:1-23:
As seen in FIG. 1, interactive network 10 uses a layered
structure that includes an information layer 100, a switch/file
server layer 200, and cache/concentrator layer 300 as well as
reception layer 401. This structure maintains active application
databases and delivers requested parts of the databases on
375

Reference

Disclosure
demand to the plurality of RS 400’s, shown in FIG. 2. As seen
in FIG. 2, cache/concentrator layer 300 includes a plurality of
cache/concentrator units 302, each or which serve a plurality of
RS 400 units over lines 301. Additionally, switch/file server
layer 200 is seen to include a server unit 205 connected to
multiple cache/concentrator units 302 over lines 201. Still
further, server unit 205 is seen to be connected to information
layer 100 and its various elements, which act as means for
producing, supplying and maintaining the network databases
and other information necessary to support network 10.
Continuing, switch/filer layer 200 is also seen to include
gateway systems 210 connected to server 205. Gateways 210
couple layer 200 to other sources of information and data; e.g.,
other computer systems. As will be appreciated by those
skilled in the art, layer 200, like layers 401 and 300, could also
include multiple servers, gateways and information layers in
the event even larger numbers of users were sought to be
served.
https://web.archive.
See, e.g., “The integration of traditional databases into scalable Web
org/web/19961107
servers. Although the primary database for the HotBot search engine is
001155/http://www.inkt custom-built for high performance, we use an integrated multiomi.com/tec
machine Informix database for tracking user preference profiles and ad
hnology.html
placement and accounting. Informix provides multi-platform parallel
database queries that fit well with the building-block model used by
Inktomi: each server has the full power of SQL transactions and we
replicate information to provide fault tolerance. The pervasive use of
dynamic HTML generation to allow every user to see a customized
page. The use of mass customization, in which we treat millions of
users individually within one framework, requires scalable computing
resources and database integration, but also requires new tools and
technology. In particular, we have developed a new form of dynamic
HTML that includes a server-side scripting language that generates
HTML on the fly based on the user profile and client browser
information. In addition to the obvious benefit of allowing users to
customize their page, this technology also enables more targeted
advertising, and use of advanced HTML features such as frames and
tables for those browsers that can support them; we are not limited to
some "least-common denominator" subset of HTML (for example to
support older browsers).”
https://web.archive.org/
web/19961107001258/
http://www.inktomi.
com/whitepap.html

See, e.g., Database access. Audience1 comes with Dynamic tags that
can access a DBMS for arbitrary persistent information and customize
the HTML tracking, using either cookies or fat URLs. Unlike other
offerings, while Audience1 supports SQL, it does not require
publishers to know SQL to access the database. This allows Inktomi
376

Reference

DUMMIES

PINKERTON

Disclosure
servers to store and recall a user's preferences for user interface and
query results presentation. More generally, Audience1 is ideal for
allowing servers to access pre-existing databases such as products,
inventory, etc. Browser targeting. Audience1 allows publishers to
exploit leading-edge HTML features (such as Netscape's frames and
Java, and Microsoft's font changes and embedded audio tags), without
frustrating users who do not have those features. Audience1's browser
targeting can be performed at various levels of detail, ranging from
tags that are easy to use, but don't provide a lot of publishing control,
to exposing the raw browser capabilities to the publisher. For
example, advertisers on HotBot are shown as progressive JPEG if the
client browser supports it, otherwise they are shown as JPEGs or GIFs
for less-capable browsers. This allows Inktomi to make the most of
each browser, rather than resorting to a least-common denominator.
Access to high performance, scalable services. Dynamic Tags make it
possible for publishers to introduce new, high performance, scalable
services, without requiring the publisher to understand the intricacies
of computing programming. For example, access to the Inktomi search
engine is encapsulated into a single Dynamic Tag, hiding the
complexity of interfacing to a parallel program such as Inktomi. In
addition, Dynamic Tags can be multi-threaded, interleaving longlatency operations such as Inktomi queries and customized content
selection (i.e. targeted advertisements). We know of no other Webbased publishing system with this capability and ease-of-use.
Publishing support hides the complexity of creating and managing
sites of dynamic Web pages, allowing sites with large amounts of
content to control the publishing process. Unlike the CGI-based tools
that are emerging, Audience1's publishing support is fault tolerant,
high performance and scales to millions of users and millions of hits
per day. In summary, Audience1 and Dynamic Tags allow a
customizable and sophisticated user-interface to Web services such as
search engine. HotBot's interface, including saved searches,
personalization, and browser targeting, would have been nearly
impossible without the simplification provided by the Audience1
toolset.”
See e.g., DUMMIES, p. 87-88 (identifying the three databases that may
be searched by the Lycos search engine: a2z directory, Lycos catalog,
and Point reviews.); id., p. 103-104 (describing the different databases
available to search with the Excite search engine, the Web, Usenet,
Classifieds, and Reviews)
PINKERTON, P., 2 (“After retrieving a document, the WebCrawler
performs three actions: it marks the document as having been
retrieved, deciphers any outbound links (href’s), and indexes the
content of the document. All of these steps involve storing
information in a database”); id., p. 2-3 (“The database handles the
377

Reference

Disclosure
persistent storage of the document metadata, the links between
documents, and the full-text index”); id., p. 5 (“The WebCrawler’s
database is comprised of two separate pieces: a full-text index and a
representation of the Web as a graph. The database is stored on disk,
and is updated as documents are added.”)

NETGRAVITY
ADSERVER HELP

See e.g., NETGRAVITY ADSERVER HELP, Choosing an Installation
Scenario: “AdServerUI Host – To manage your ads and ad schedules
you install the AdServerUI, which provides a Web interface for
administering the AdServer database. The machine on which the
AdServerUI resides is called the AdServerUI host. Content Host –
The content host is the machine that runs your Web server and
contains your Web content tree. Your site may have multiple content
hosts. Though described above as separate, the content host and the
AdServerUI host can, in fact, be the same machine. In other words,
all AdServer components may be installed on the same host. Or, you
may choose to host them on separate machines . . . Though depicted in
the above diagram as separate, the content host and the AsServerUI
host can be the same machine.”); id., Configuration Directives, p. 5
(“When you restart AdServer, it copies the database from the
DatabaseStageDir to the DatabaseDir, and begins serving ads from
this new database.”); id., Configuring Your Content Server (“Your
content server is the HTTP server that you use to serve your Web
content.”); id., Dynamic Ad Placement: Overview (“To serve an ad
dynamically means that whenever an ad needs to be shown, the
content server asks AdServer which ad to display at that exact
moment.”)
See e.g., ABOUT NETGRAVITY ADSERVER, Getting Started, p. 2
(“AdManager writes to the AdServer database, recording ad and
scheduling information. . . . The content server is the Web server that
serves your site’s content pages.”); see also id., Installing AdServer.
FLYNN, p. 2-3 (“In the NetGravity model, advertisers can store their
ads on their own server or the site’s server.”)
See e.g., DEDRICK 1994, p. 55 (“Typical consumption devices are
personal home computers that are connected to an electronic content
distribution network via transport technologies such as cable, satellite,
ISDN, POTS, and wireless . . .”); id., p. 56 (“Fig. 1 shows an end-toend high-level view of a content distribution network. This network
connects content authors of ‘rich media’ advertisements with business
and home content consumers.”); id. (“Fig. 1 shows the network
connections that will allow bi-directional communication between
authors and consumers, consumers and authors, etc.”); id., p. 57 (“The
model proposed for dissemination of interactive electronic
advertisements is through a series of cooperating local electronic
yellow pages services, each spanning a specified region (with

ABOUT NETGRAVITY
ADSERVER

FLYNN
DEDRICK 1994

378

Reference

DEDRICK 1995

GALLAGHER

Disclosure
potential for overlapping regions). Additionally, these local yellow
pages servers also have connectivity with larger regional, national, and
global electronic yellow pages services. To enable electronic
advertising to subsidize the consumption of electronic content, these
yellow pages services are also integrated with a variety of related
services.”); id., p. 59 (“All consumers having access to the local
electronic yellow pages can search these yellow pages, compare prices
when multiple listings of similar service offerings exist, and
automatically schedule an appointment with a service provider.”); id.,
p. 59 (“Object-oriented database management is one of the core
required technologies of an electronic yellow pages mechanism. Such
a DBMS must have distributed data management capabilities to deal
with electronic advertisements existing across multiple regions.”); id.,
p. 60 (“the content distribution architecture is largely client-server
oriented, using large hard-disk intensive network servers to hold
terabytes of electronic content.”); id., p. 61 (“Back-channel
capabilities enable a client consumption device to send requests to the
electronic content distribution network servers and also to other
network clients.”); id., p. 62 (“Specifically, the currently suggested
attribute extension list as follows: . . . Dynamic (e.g., hypertext) links
to associated objects, residing on both local and remote servers.”)
See e.g., DEDRICK 1995, p. 42 (An end-to-end electronic content
distribution network connects connect authors of rich-media
advertising with business and home content consumers. . . . Network
connections must provide connectivity that will allow bidirectional
communication between authors and consumers. In addition, the endto-end distribution network must include intermediate content
repositories.”); id., p. 43 (“To enable electronic advertising to
subsidize content, these yellow pages services are also integrated with
related electronic services, including commerce financial
clearinghouses, content databases, authors, and content delivery to
consumers.”); id., p. 43, Fig. 1; id., p. 44 ( “Distributed database
management is one of the core required technologies of an electronic
yellow pages mechanism.”); id., p. 46 (“All information on a profile
device is protected by encryption and made available to the consumer
only when the profile device is plugged into a consumption device and
the consumer has entered the correct decryption password or personal
identification number (PIN). Second, using a portable hardware-based
device as a repository of consumers’ personal profiles lets consumers
plug into the content distribution network through any device at work
or at home.”)
See e.g., GALLAGHER, p. 5 (“The architecture required to implement
the basic version of the model consists of two parts: data structure to
represent user profiles and target audience profiles, and an algorithm
to select banner advertisements to display to a user.”); id., p. 6
379

Reference

Lycos, Inc. Registration
Statement No. 333-354,
dated April 3, 1996
(“LYCOS PROSPECUS”),
produced at GOOGWRD-00872476GOOG-WRD00872549

Disclosure
(“Operationally, profiles for ideal or acceptable users can be
maintained in a relational database structure. In the case of ideal
profiles, a table can be defined in which each row describes the ideal
target audience member for each advertisement..”); id., p. 8, Fig. 4.
See LYCOS PROSPECTUS at GOOG-WRD-00872482:

Id. at GOOG-WRD-00872484:

Id. at GOOG-WRD-00872497-498:

Id. at GOOG-WRD-00872499:

380

Reference

Disclosure

Id. at GOOG-WRD-00872502:

Lycos, Inc. Form S-1
Registration Statement,
dated February 14,
1996 (“LYCOS S-1”),
produced at GOOGWRD-00872550GOOG-WRD00872923

See LYCOS S-1 at GOOG-WRD-00872558:

Id. at GOOG-WRD-00872560:

Id. at GOOG-WRD-00872573:

381

Reference

Disclosure

Id. at GOOG-WRD-00872575:

Id. at GOOG-WRD-00872578:

Excite, Inc. SB-2
Registration Statement
No. 333-2328-LA,
March 11, 1996
(“Excite SB-2”)
produced at GOOGWRD-00872006GOOG-WRD00872094

Id. at GOOG-WRD-0087209.

Id. at GOOG-WRD-0087209.
382

Reference

Disclosure

Id. at GOOG-WRD-00872010.

Id. at GOOG-WRD-00872011.

383

Reference

Disclosure

Id. at GOOG-WRD-00872017-18.

Id. at GOOG-WRD-00872038.

Id. at GOOG-WRD-00872043.

384

Reference

Disclosure

Id. at GOOG-WRD-00872044.

Id.

Id. at GOOG-WRD-00872035.

385

Reference

Disclosure

Id.

Id. at GOOG-WRD-00872037.

Id. at GOOG-WRD-00872038.

Id.

Id. at GOOG-WRD-00872039.

386

Reference

Disclosure

Id. at GOOG-WRD-00872042.

Id. at GOOG-WRD-00872046.

387

Reference

Disclosure

Id. at GOOG-WRD-00872047-48.

Id. at GOOG-WRD-00872048.

Id. at GOOG-WRD-00872049.
Excite, Inc. Prospectus,
dated April 3, 1996
(“Excite Prospectus”)
produced at GOOGWRD-00871928GOOGL-WRD00872005

Id. at GOOG-WRD-00871929.

388

Reference

Disclosure

Id. at GOOG-WRD-00871929.

Id. at GOOG-WRD-00871930.

389

Reference

Disclosure

Id. at GOOG-WRD-00871931.

Id. at GOOG-WRD-00871937-38.

390

Reference

Disclosure

Id. at GOOG-WRD-00871958.

Id. at GOOG-WRD-00871963.

Id. at GOOG-WRD-00871964.

Id.

391

Reference

Disclosure

Id. at GOOG-WRD-00871955.

Id.

Id. at GOOG-WRD-00871957.

Id. at GOOG-WRD-00871958.

392

Reference

Disclosure

Id.

Id. at GOOG-WRD-00871959.

Id. at GOOG-WRD-00871962.

393

Reference

Disclosure

Id. at GOOG-WRD-00871966.

Id. at GOOG-WRD-00871967-68.

394

Reference

Disclosure

Id. at GOOG-WRD-00871968.

Id. at GOOG-WRD-00871969.

395

Reference
InfoSeek Corporation
S-1 Registration
Statement No. 3334142, Amendment No.
1, dated May 3, 1996
(“InfoSeek S-1”)
produced at GOOGWRD-00872371GOOG-WRD00872464

Disclosure

Id. at GOOG-WRD-00872375.

Id. at GOOG-WRD-00872403.

396

Reference

Disclosure

Id.

Id. at GOOG-WRD-00872404.

Id. at GOOG-WRD-00872404-05.

Id. at GOOG-WRD-00872405.

397

Reference

Disclosure

Id. at GOOG-WRD-00872406.

Id.

398

Reference

Disclosure
Id. at GOOG-WRD-00872408.

Id. at GOOG-WRD-00872408-09.

Id. at GOOG-WRD-00872409-10.

Id. at GOOG-WRD-00872410.

Id. at GOOG-WRD-00872411.

399

Reference

Disclosure

Id. at GOOG-WRD-00872413.
Yahoo Prospectus
Registration Statement
No. 333-2142, dated
April 12, 1996 (“Yahoo
Prospectus”) produced
at GOOG-WRD00874251-GOOGWRD-00874328

Id. at GOOG-WRD-00874279.

Id. at GOOG-WRD-00874280.

400

Reference

Disclosure
Id. at GOOG-WRD-00874281.

Id. at GOOG-WRD-00874282.

Id. at GOOG-WRD-00874287.

Id. at GOOG-WRD-00874290.

Id. at GOOG-WRD-00874291.

401

Reference
Yahoo Form SB-2
Registration Statement
No. 333-2142, dated
March 7, 1996 (“Yahoo
Form SB-2”) produced
at GOOG-WRD00874329-GOOGWRD-00874418

Disclosure

Id. at GOOG-WRD-00874357.

Id. at GOOG-WRD-00874358.

Id. at GOOG-WRD-00874359.

402

Reference

Disclosure

Id. at GOOG-WRD-00874360.

Id. at GOOG-WRD-00874365.

Id. at GOOG-WRD-00874368.

Id. at GOOG-WRD-00874368-69.

403

Reference
Disclosure
Open Text Form F-1
Registration Statement
No. 33-98858, dated
November 1, 1995
(“Open Text Form F1”) produced at GOOGWRD-00873727GOOG-WRD00873878

Id. at GOOG-WRD-00873603.

404

Reference

Disclosure

405

Reference

Disclosure

Id. at GOOG-WRD-00873633-35.

Id. at GOOG-WRD-00873639.

Id. at GOOG-WRD-00873640.
406

Reference

Disclosure

Id. at GOOG-WRD-00873641.

Id. at GOOG-WRD-00873642-43.

Id. at GOOG-WRD-00873650.

407

Reference

Disclosure
Id. at GOOG-WRD-00873675.

Id. at GOOG-WRD-00873676.

Id. at GOOG-WRD-00873677.
Open Prospectus, dated
January 23, 1996
(“Open Text
Prospectus”) produced
at OT03652-3758

Id. at OT03653.

408

Reference

Disclosure

409

Reference

Disclosure

410

Reference

Disclosure

Id. at OT03689-91.

Id. at OT03695.

411

Reference

Disclosure

Id. at OT03696.

Id. at OT03697.

Id. at OT03698.

Id. at OT03735.
412

Reference

Disclosure

Id. at OT03736.

Id. at OT03737.

413