Campbell et al v. Facebook Inc.

Filing 179

Joint Notice of Refiling of Documents Accompanying Class Certification Briefs and Evidentiary Objections. (Attachments: # 1 Exhibit 1 (Redacted), # 2 Exhibit 2 (Redacted), # 3 Exhibit 3 (Redacted), # 4 Exhibit 4 (Redacted), # 5 Exhibit 5 (Redacted), # 6 Exhibit 6 (Redacted), # 7 Exhibit 7 (Redacted), # 8 Exhibit 8 (Unredacted), # 9 Exhibit 9 (Unredacted), # 10 Exhibit 10 (Redacted), # 11 Exhibit 11 (Redacted), # 12 Exhibit 12 (Redacted), # 13 Exhibit 13 (Redacted), # 14 Exhibit 14 (Redacted), # 15 Exhibit 15 (Redacted), # 16 Exhibit 16 (Redacted), # 17 Exhibit 17 (Redacted), # 18 Exhibit 18 (Redacted), # 19 Exhibit 19 (Unredacted), # 20 Exhibit 20 (Redacted), # 21 Exhibit 21 (Redacted), # 22 Exhibit 22 (Unredacted), # 23 Exhibit 23 (Redacted), # 24 Exhibit 24 (Redacted), # 25 Exhibit 25 (Unredacted))(Chorba, Christopher) (Filed on 3/28/2016) Modified on 3/29/2016 (kcS, COURT STAFF).

EXHIBIT 8

From:

Sent:
To:

Subject:
Attachments:

Max Eulenstein
Sunday, March 25, 2012 8:47 PM
Christopher Plambeck; Jeff Mccombs
RE: Like constraint
Like Distribution.pdf

Attached is a graph showing the global distribution of likes. The global median is approximately 20 likes and the global
average is approximately 90 likes. This varies by country and by the type of pages included (in the U.S., the median is 28
and the average is 102).
It is an acknowledged problem that a shortage of likes is limiting the number of users that can be targeted by their
interests and thereby affecting revenue. The targeting team, led by Antonio, has been attempting to address this
problem for the past 5 months - it's been the major focus of the team. The first effort was an aggressive expansion of
the PY~v1L (Pages you ~v1ay Like) Unit. The unit has been successful by some measures, including a 1% revenue gain
against the backtest, but it has not significantly shifted the distribution of users with likes partially because low-like users
have the lowest conversion rates with the unit. The second effort is to assign users likes based on the likes of their
closest friends and other signal we may have (e.g. user writes). This has proved to be initially successful, but before
expanding efforts here, the team is adding keyword features to our click-prediction model so that as we propagate likes
more aggressively we don't flood the system with bad-signal and hurt advertiser value.
If you're interested in the details, I can definitely get the targeting roadmap from Antonio and pass it along.
-----Original Message----From: Christopher Plambeck
Sent: Sunday, March 25, 2012 1:40 PM
To: Jeff Mccombs; Max Eulenstein
Subject: RE: Like constraint
+Max, who has some data on distribution of likes across users - it is highly skewed with something like the mean user
having liked 50 pages but the median users has only 10 (but Max should be able to provide the actual numbers). That's
one reason it can often be costly to use the page post unit - fans of page X are very likely also fans of Page Y, and so for
some users, we may actually be more supply constrained.
The Pages You May Like test was designed to group likes - Max is also doing some work around the impact of that, but
for the next set of tests, we are shifting that to use the PYML unit to get fans for newer local pages in order to increase
their chances of staying active. The Targeting team has also done I is doing work around using your friends targeting
interests as a targeting signal for users that lack direct likes and interests, and there are probably other things beyond
those that the targeting team is doing - Max may be able to discuss those or could get you Antonio and team's roadmap.
What's the broader context?
Chris
-----Original Message----From: Jeff Mccombs
Sent: Saturday, March 24, 2012 12:18 PM
To: Christopher Plambeck
Subject: Like constraint

Highly Confidential

FB000014365

Do you have any data on the issue of being like-constrained from a targeting perspective and what efforts are underway
to address this? Thanks!
Sent from my iPhone

2

Highly Confidential

FB000014366