I/P Engine, Inc. v. AOL, Inc. et al

Filing 528

Declaration re 527 Reply to Response to Motion, of Joshua L. Sohn in Support of the Reply in Support of Defendants' Motion for Summary Judgment by AOL Inc., Gannett Company, Inc., Google Inc., IAC Search & Media, Inc., Target Corporation. (Attachments: # 1 Exhibit 34, # 2 Exhibit 35, # 3 Exhibit 36, # 4 Exhibit 37, # 5 Exhibit 38)(Noona, Stephen)

EXHIBIT 38

Confidential Pursuant to Protective Order

Page 1
1

UNITED STATES DISTRICT COURT

2

EASTERN DISTRICT OF VIRGINIA

3

NORFOLK DIVISION

4

-------------------------x
I/P ENGINE, INC.,

5
Plaintiff,
6
v.

Civil Action No. 2:11-cv-512

7
GOOGLE INC., et al.,
8
Defendants.
9

-------------------------x

10
11

CONFIDENTIAL PURSUANT TO PROTECTIVE ORDER

12
13

Videotaped Deposition of DONALD M. KOSAK

14

Washington, D.C.

15

Thursday, May 31, 2012

16

9:04 a.m.

17
18
19
20
21
22

Reported by:

Amy E. Sikora, RPR, CRR, CSR-NY, CLR

23
24
25

Job No. CS397174
Veritext Corporate Services
800-567-8658

973-410-4040

Confidential Pursuant to Protective Order

Page 213
1

Do you know what your early

2

information filtering techniques were?

3

A.

I don't know what they're referring to

4

there.

5

document and this is the first that I've seen

6

this.

7
8

I -- I had no part in creating this

Q.

Well, you are listed as a consultant

in Innovate/Protect; right?

9

A.

I don't know.

10

Q.

Take a look at the last page.

11

A.

Hmm.

12

Q.

Well, did you adopt -- adapt any of

I see my name on the last page.

13

your early information filtering techniques to

14

apply to search systems?

15

MS. ALBERT:

16

foundation.

17

A.

Objection.

No

Asked and answered.
Certainly we acquired a large number

18

of techniques that we developed in working with

19

content and working with information filtering

20

systems.

21

have used some of those techniques and applied

22

some of those techniques to the domain of search

23

systems or information retrieval.

24
25

Q.

And it's a true statement that we may

Can you think of any techniques that

you did adapt and apply to search?
Veritext Corporate Services
800-567-8658

973-410-4040

Confidential Pursuant to Protective Order

Page 214
1

A.

In -- in general or --

2

Q.

Yeah.

3

A.

At my time at Lycos.

4

very broad.

5

what, 14 years?

And you're asking me if I ever

6

did something.

I don't know if I can answer that

7

in any other way than -- than, I don't know.

The question's

I mean, it covers the scope of,

8

Q.

All right.

So let's cabinet it in

9

time, then.

10

A.

Okay.

11

Q.

Prior to December 1998, did you ever

12

adapt any techniques from information filtering

13

to search systems that you were creating?

14
15

A.

We certainly used various techniques

in some of those research projects that we built.

16

Q.

And what techniques were those?

17

A.

Various ways of parsing documents.

18

Various ways of stemming -- stemming is a

19

technical term.

20

Just an entire litany of -- of techniques.

21
22
23
24
25

Q.

Different linguistic analysis.

Any techniques related to

collaborative filtering?
A.

Certainly some of the techniques had

something to do with collaborative filtering.
Q.

Do you remember which of the
Veritext Corporate Services

800-567-8658

973-410-4040

Confidential Pursuant to Protective Order

Page 215
1

techniques had to do with collaborative filtering

2

that you used in search --

3

A.

I don't know.

4

Q.

-- or adapted from the early

5

information filtering techniques?

6

MS. ALBERT:

7

A.

Objection.

It's difficult for me to answer

8

because we had a lot of different research

9

projects, and many of them were on around ways of

10

improving the search experience.

11

you know, 1998 that we're talking about.

12

recall exactly which experiments had which pieces

13

in it.

14

And this was,
I don't

You know, these are not things that I

15

spent hours every day on the experiments.

16

were things that I parceled off to people to run

17

tests and get results back to me.

18

possibly a few hours of experience on some of

19

them and, you know, maybe a day of experience on

20

another.

21

indelibly etched in my memory.

22

Q.

They

So I've got

They're not things that are going to be

See, now I'm confused because earlier

23

today you said that you and Mr. Lang were the

24

only ones that were working on the search part of

25

the project?
Veritext Corporate Services
800-567-8658

973-410-4040

Confidential Pursuant to Protective Order

Page 216
1

A.

From management.

You asked me a

2

question about managing, and I answered that Ken

3

and I were the only ones that oversaw that

4

information.

5

Q.

So who else worked --

6

A.

From the management standpoint.

7

Q.

Sorry.

8

A.

Sorry to interrupt.

9

Q.

Are you done?

10

A.

I am finished.

11

Q.

Who else worked on search, those

12

search research projects that incorporated

13

content-based- and collaborative filtering

14

besides you and Mr. Lang?

15

A.

Well, there were two teams that did

16

some of the experiments.

17

group.

18

precisely at that particular time period that

19

research group was.

20

called the Lycos advanced product development

21

group that basically built prototypes, not

22

finished products.

23

would be the direct ones responsible.

24
25

Q.

There was the research

I don't recall what -- what size

There was another group

So those two different teams

Do you remember any individuals in the

research group?
Veritext Corporate Services
800-567-8658

973-410-4040

Confidential Pursuant to Protective Order

Page 253
1

Q.

All right.

Then tell me what happens.

2

I make a demand search.

3

There's no wire.

4
5
6
7
8
9

A.

I put in a query.

What happens?

We use the content-based side of the

filter to generate a list of results.
Q.

Okay.

After the content-based filter

generates a list of results, what happens next?
A.

In this prototype that we're

discussing, at that point the results were

10

displayed to the user; in this case, a

11

researcher.

12

some of the results.

13

the results, we tabulated that as a feedback into

14

the system.

15

that result was clicked.

16

the pool of information that we used on the

17

collaborative side.

18

Q.

That person may or may not click on
If they clicked on some of

You know, this result was clicked,
And modified or created

Well, how did it modify or -- how did

19

it modify the pool of information that you used

20

on the collaborative side, if the person already

21

received the results of that query?

22

A.

That would be for the next person

23

coming through asking for that query.

So if the

24

next person came through and typed in a query,

25

the same query, for example, yeah.
Veritext Corporate Services
800-567-8658

973-410-4040

Confidential Pursuant to Protective Order

Page 254
1

Q.

So the first time a user made a demand

2

search and there wasn't a wire for it, the

3

collaborative part of the algorithm did not kick

4

in?

5
6

MS. ALBERT:
A.

Objection.

Well, there -- there were instances

7

that we could use the information that we might

8

have on one of the results that came up to make a

9

determination as to whether, you know, the

10

ranking of that result should be moved up or

11

down.

12

Q.

So tell me how that worked.

13

A.

In our prototype system, when -- when

14

you had multiple queries coming through, the

15

queries didn't necessarily have to generate the

16

exact same result set.

17

document that's in common between those result

18

sets.

19

semi-related query (indicating) clicked on that

20

document, might make that document rise up

21

numerically its score higher.

22

behavior might influence other queries.

23

But there might be a

The fact that somebody at this

Q.

So that pattern of

How did you know that a query was

24

semi-related as opposed to -- if it wasn't

25

identical, how did you know that it was
Veritext Corporate Services
800-567-8658

973-410-4040

Confidential Pursuant to Protective Order

Page 255
1

semi-related?

2

A.

Well, in my example I'm talking about

3

a case where the same result or the same document

4

was shown in both queries.

5

conclusion that there was a relationship between

6

the queries because they returned an instance of

7

the same document.

8

a very finite example here.

9

Q.

So I'm drawing the

And I'm kind of talking about

And then when you used -- when you

10

used that information where somebody had clicked

11

on the same document that showed up in the two

12

queries, what happened next?

13

A.

I don't know.

I could speculate.

I

14

mean, this -- this prototype that we're talking

15

about is a collection of different things and

16

it's, what?

17

hundreds of different corner cases.

18

it did something for those corner cases.

19

remember every single corner case a decade after

20

it was done.

21

Q.

22

one.

23

12 years ago.

We could walk through
I know that
I can't

Well, let's not go for every single

Let's just talk about the ones you

remember.

24

A.

25

Well, okay.

On those corner cases, I

don't know.
Veritext Corporate Services
800-567-8658

973-410-4040

Confidential Pursuant to Protective Order

Page 256
1

Q.

What do you mean by "corner cases"?

2

A.

When you get down to the detail level

3

of how some particular co-occurrence of different

4

things by different users for different document

5

sets for hypothetical queries, I'm not going to

6

be able to answer questions like that 10 years

7

after, you know, this was done.

8
9

Again, to remind you, I didn't build
these prototypes.

I helped collaborate with the

10

design of these things with Ken Lang who directed

11

the building of these prototypes.

12

much more familiar with the prototypes than

13

myself.

14

Q.

He would be

So when the prototypes were built, did

15

the builders have any leeway on how they were

16

implementing things or were they told exactly

17

what to do?

18

A.

I don't know how Ken directed them.

19

did not attend any of Ken's staff meetings or,

20

you know, research group meetings.

21

know, busy.

22

and myself.

23

Q.

I

24
25

I was, you

The collaboration was between Ken

The people that were busy building the

system, what kind of backgrounds did they have?
A.

Are you talking about the people who
Veritext Corporate Services

800-567-8658

973-410-4040

Confidential Pursuant to Protective Order

Page 257
1
2
3
4

reported to Ken Lang?
Q.

Yeah.

Who actually did the building

of the system.
A.

Well, I don't know who Ken had

5

building the various things or carrying out his

6

projects.

7

were Ph.D.'s.

8

were in the process of getting Ph.D.'s, and he

9

had some really hard core software engineers.

10

Ken's staff had a bunch of people who
He had a handful of people who

Most of the time the research group

11

had Ph.D. guys, and the really hard core

12

engineers were in the prototyping advanced

13

development group.

14

overlap.

15

he could mix them together how he wanted.

16

don't know which combinations of staff he used.

17

Q.

And sometimes there was some

And the reason Ken had both of them is
So I

It sounds like the people had a pretty

18

high level of educational background, though,

19

that were working on implementing this; is that

20

right?

21
22
23

A.

I think that characterizes most of the

employees then, yes.
Q.

And would you say that was

24

representative of the skill level of people

25

working in this field at that time in 1998?
Veritext Corporate Services
800-567-8658

973-410-4040