Facebook, Inc.

Patent Trials and Appeals Board

Jan 29, 2021

2020003121 (P.T.A.B. Jan. 29, 2021)

UNITED STATES PATENT AND TRADEMARK OFFICE
UNITED STATES DEPARTMENT OF COMMERCE
United States Patent and Trademark Office
Address: COMMISSIONER FOR PATENTS
P.O. Box 1450
Alexandria, Virginia 22313-1450
www.uspto.gov
APPLICATION NO. FILING DATE FIRST NAMED INVENTOR ATTORNEY DOCKET NO. CONFIRMATION NO.
15/174,865 06/06/2016 Robert Oliver Burns Zeldin 108282.020540 6620
169063 7590 01/29/2021
BakerHostetler / Facebook
Cira Centre 12th Floor
2929 Arch Street
Philadelphia, PA 19104-2891
EXAMINER
EVANS, KIMBERLY L
ART UNIT PAPER NUMBER
3629
NOTIFICATION DATE DELIVERY MODE
01/29/2021 ELECTRONIC
Please find below and/or attached an Office communication concerning this application or proceeding.
The time period for reply, if any, is set in the attached communication.
Notice of the Office communication was sent electronically on above-indicated "Notification Date" to the
following e-mail address(es):
eofficemonitor@bakerlaw.com
PTOL-90A (Rev. 04/07)
UNITED STATES PATENT AND TRADEMARK OFFICE
____________

BEFORE THE PATENT TRIAL AND APPEAL BOARD
____________

Ex parte ROBERT OLIVER BURNS ZELDIN,
NATHAN JOHN DAVIS, ANAND SUMATILAL BHALGAT,
HARSH DOSHI, and HAO SONG
____________

Appeal 2020-003121
Application 15/174,8651
Technology Center 3600

____________

Before JAMES P. CALVE, BRUCE T. WIEDER, and
BRADLEY B. BAYAT, Administrative Patent Judges.

WIEDER, Administrative Patent Judge.

DECISION ON APPEAL
This is a decision on appeal under 35 U.S.C. § 134 from the
Examiner’s rejection of claims 1–21. We have jurisdiction under
35 U.S.C. § 6(b).
We REVERSE.

1 We use the word “Appellant” to refer to “applicant” as defined in
37 C.F.R. § 1.42. Appellant identifies the real party in interest as Facebook
Technologies, LLC. (Appeal Br. 1.)
Appeal 2020-003121
Application 15/174,865

2
CLAIMED SUBJECT MATTER
Appellant’s invention relates “to selecting content to online system
users based on actions by the users at least a reasonable amount of time after
presentation of the content.” (Spec. ¶ 1.)
Claims 1 and 12 are the independent claims on appeal. Claim 1 is
illustrative. It recites (emphasis added):
1. A method comprising:
receiving content items at an online system for
presentation to users of the online system;
retrieving models maintained by the online system for
determining likelihoods of users performing one or more
interactions with presented content items within a threshold
amount of time after presentation of the presented content items
based on characteristics of the users and characteristics of
content items;
determining weights associated with each of the retrieved
models based on prior actions by users after presentation of
content items to users of the online system;
generating a model for determining a latent metric
describing user actions occurring after the threshold amount of
time after presentation of content items by applying the weights
to the retrieved models associated with the weights and
combining the retrieved models after application of the weights;
identifying an opportunity to present one or more content
items to the user;
identifying content items eligible for presentation to the
user from the received content items;
determining the latent metric describing user actions
occurring after the threshold amount of time after presentation of
an identified content item for each identified content item using
the model for determining the latent metric describing user
actions occurring after the threshold amount of time after
presentation of content items;
selecting a content item of the identified content items for
presentation to the user based on the determined latent metrics;
and
Appeal 2020-003121
Application 15/174,865

3
providing the selected content item to a client device for
presentation to the user.

REJECTIONS2
Claims 10 and 21 are rejected under 35 U.S.C. § 112(b) as indefinite
for failing to particularly point out and distinctly claim the subject matter
which the inventor or a joint inventor regards as the invention.
Claims 1, 2, 4–6, 10–13, 15–17, and 21 are rejected under 35 U.S.C.
§ 103 as unpatentable in view of Zhang (US 2014/0068011 A1, pub.
Mar. 6, 2014) and Phil Blunsom, Hidden Markov Models, Aug. 19, 2004,
(available at http://tka4.org/materials/lib/Articles-
Books/Speech%20Recognition/hmm-tutorial.pdf (last visited Jan. 27, 2021)
(hereinafter “Blunsom”)).
Claims 3 and 14 are rejected under 35 U.S.C. § 103 as unpatentable in
view of Zhang, Blunsom, and El-Rafei (US 2009/0018983 A1, pub.
Jan. 15, 2009).
Claims 7–9 and 18–20 are rejected under 35 U.S.C. § 103 as
unpatentable in view of Zhang, Blunsom, and Karande
(US 2015/0206170 A1, pub. July 23, 2015).

2 The Answer states that “[p]er the Advisory Action dated 9/11/19, the
rejection of claims 10 and 21 under 35 U.S.C. 112(b) has been withdrawn.”
(Answer 4.) However, the Advisory Action maintains the rejection of
claims 10 and 21 under § 112(b), and the § 112(b) rejection is addressed in
the Answer. (See id. at 4–6.) The Advisory Action withdraws the § 101
rejection of claims 12–21. (See Advisory Action dated Sept. 11, 2019.)
Moreover, the § 112(b) rejection is argued in the appeal briefs and the § 101
rejection is not argued. Therefore, we treat as erroneous the statement in the
Answer that the § 112(b) rejection is withdrawn, and instead treat the § 101
rejection as withdrawn.
Appeal 2020-003121
Application 15/174,865

4
ANALYSIS
The § 112(b) rejection of dependent claims 10 and 21
Claim 10 recites (paragraphing and emphasis added):
10. The method of claim 1, wherein a model maintained by
the online system for determining a likelihood of users
performing an interaction with a presented content item within
the threshold amount of time after presentation of the presented
content items is selected from a group consisting of:
a model determining a likelihood of users accessing
the presented content item,
a model determining a likelihood of users
performing a specific interaction with the presented
content item,
a model determining a likelihood of users
performing a specific interaction with an object associated
with the presented content item,
a model determining an amount of time users will
view the presented content item,
and any combination thereof.

The Examiner rejects claim 10 because the “Examiner was
unable to find support, either by way of example or a clear definition
in the specification for the conjunctive interpretation that appellant is
claiming for selecting a model.” (Answer 5.) Specifically, the
Examiner finds that
Appellant’s disclosure at ¶6 teaches, “... Example models
maintained by the online system include: a model
determining a likelihood of a user accessing content item
presented to the user, a model determining a likelihood of
the user performing a specific interaction with a content
item presented to the user (e.g., expressing a preference
for the content item, sharing the content item with another
user, commenting on the content item), a model
determining a likelihood of the user performing a specific
interaction with an object (e.g., a page, a user, etc.)
Appeal 2020-003121
Application 15/174,865

5
associated with a content item presented to the user, a
model determining an amount of time the user will view a
content item presented to the user, or models predicting
any other suitable interaction with a content item
presented to the user. The description provided in the
specification fails to provide the conjunctive interpretation
“and” suggested by appellant inclusive of all models and
combinations.

(Id. at 5–6.)
Appellant argues that “claims 10 and 21 recite an acceptable form of
Markush grouping (claim construction: ‘selecting from a group consisting of
A, B, and C’).” (Reply Br. 3.)
“A Markush group is a listing of specified alternatives of a group in a
patent claim, typically expressed in the form: a member selected from the
group consisting of A, B, and C.” Abbott Labs. v. Baxter Pharm. Prods.,
Inc., 334 F.3d 1274, 1280 (Fed. Cir. 2003). A Markush group should be
closed, i.e., it should be characterized with the phrase “consisting of” rather
than, e.g., comprising. Id.
Claim 10 recites “a model maintained by the online system . . . is
selected from a group consisting of: a model . . . , a model . . . , a model
. . . , a model . . . , and any combination thereof.” Claim 10 is in proper
Markush form. We do not agree that claim 10 is indefinite under § 112(b),
and we will reverse this rejection of claim 10 and of claim 21, which
contains similar language.3

3 The Examiner suggests that these claims, properly construed, may not have
sufficient written description support under § 112(a). (See Answer 5–6.)
But the Examiner does not make such a rejection, and we do not address this
question in the first instance.
Appeal 2020-003121
Application 15/174,865

6
The § 103 rejections
Obviousness is a legal conclusion involving a determination of
underlying facts.
Under § 103, the scope and content of the prior art are to be
determined; differences between the prior art and the claims at
issue are to be ascertained; and the level of ordinary skill in the
pertinent art resolved. Against this background, the obviousness
or nonobviousness of the subject matter is determined. Such
secondary considerations as commercial success, long felt but
unsolved needs, failure of others, etc., might be utilized to give
light to the circumstances surrounding the origin of the subject
matter sought to be patented.

KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398, 406 (2007) (quoting Graham v.
John Deere Co. of Kansas City, 383 U.S. 1, 17–18 (1966)).
With regard to the scope and content of the prior art, the Examiner
finds that Zhang teaches
content selection module 162 may use the various data in client
identifier data 190 and content database 122 as inputs to a
predictive model, such as a linear regression model. For
example, content selection module 162 may generate a pCTR
[(predicted click-through rate)] or pCVR [(predicted conversion
rate4)] for third-party content eligible for selection for client
device 102. Content selection service 198 may use such a metric
in the content selection process. . . . Hence, Zhang teaches the
intended function of appellant’s weights associated with models
based on retrieved user actions.

(Answer 8–9 (citing Zhang ¶ 30).)

4 “Generally, a conversion refers to any specified type of online action
performed by a user after being presented the third-party content. For
example, a conversion may correspond to the user clicking on the third-party
content and making a purchase at the third-party content provider’s website,
signing up for an online newsletter, downloading software, etc.” (Zhang
¶ 19.)
Appeal 2020-003121
Application 15/174,865

7
The Examiner also finds that Blunsom teaches
various models including weighted models (equations 31, 32) for
generating sequences (log all the probability values and then add
values (equation 31) …using scaling coefficients (weights) that
keep the probability values ...that are dependent on t
(equation 32) ...-pg 7) that can be characterized by an underlying
process, hence, the intended function of appellant’s weighted
models (applying weights to models ... combing [sic] the
retrieved models).

(Id. at 11.)
The Examiner then determines that it would have been obvious “to
modify the method/system for predicting content performance of Zhang with
the Hidden Markov Models (statistical tools) as taught by Blunsom since it
allows for modeling a wide range of time series data (Abstract, Introduction
lines 1, 2, Figs 1-6, pgs 3-7).” (Id. at 11.)
Appellant argues that Zhang
determines whether an indication of a click was received in a
certain amount of time, which is distinct from applying weights
to received models and combining the retrieved models.
Similarly, Blunsom’s use of a Hidden Markov Model does not
generate a model by “applying the weights to the retrieved
models associated with the weights and combining the retrieved
models after application of the weights,” as claimed.

(Reply Br. 7–8.)
Zhang “generally relates to the selection of content for a client
device.” (Zhang ¶ 2.) More particularly, Zhang teaches using “topical
interest categories associated with [a] client identifier . . . as inputs to a
prediction model to predict the likelihood of an online action occurring as a
result of third-party content being selected.” (Id., Abstract.)
Appeal 2020-003121
Application 15/174,865

8
Blunsom relates to the use of Hidden Markov Models (HMMs). The
HMM “is a powerful statistical tool for modeling generative sequences that
can be characterised by an underlying process generating an observable
sequence. HMMs have found application in many areas interested in signal
processing . . . .” (Blunsom 1.)
Blunsom teaches that “[g]iven a HMM, and a sequence of
observations, we’d like to be able to compute . . . the probability of the
observation sequence given a model.” (Id. at 3.) One approach “is to
recognize that many redundant calculations would be made by direct[]
evaluat[ion].” (Id.) Blunsom implements a “cache [of calculations] as a
trellis of states at each time step, [and] calculat[es] the cached valued . . . for
each state as a sum over all states at the previous time step.” Blunsom uses a
“forward probability variable” to “work through the trellis filling in . . .
values.” (Id.) “The algorithm for this process is called the forward
algorithm.” (Id.)
Blunsom explains that “[w]hen implementing a HMM, floating-point
underflow is a significant problem.” (Id. at 7.) And with regard to the
forward algorithm, “[t]he most common solution to this problem is to use
scaling coefficients that keep the probability values in the dynamic range of
the [computing] machine, and that are dependent only on t [(time)].” (Id.)
We agree with the Examiner that Appellant’s Specification “does not
distinctly describe any software or algorithm specifying the criteria
regarding how the weights are determined and/or applied to the retrieved
models, and subsequently combined.” (Answer 13.) But claim 1 itself
recites both “determining [the] weights associated with each of the retrieved
models based on prior actions by users” and “applying the weights to the
Appeal 2020-003121
Application 15/174,865

9
retrieved models associated with the weights and combining the retrieved
models after application of the weights.” (See Claim 1.)
The Supreme Court in KSR Int’l Co. v. Teleflex Inc., 550
U.S. 398 . . . (2007), explained that, “because inventions in most,
if not all, instances rely upon building blocks long since
uncovered, and claimed discoveries almost of necessity will be
combinations of what, in some sense, is already known,” “it can
be important to identify a reason that would have prompted a
person of ordinary skill in the relevant field to combine the
elements in the way the claimed new invention does.” Id. at 418–
19.

Personal Web Techs., LLC v. Apple, Inc., 848 F.3d 987, 991–92 (Fed. Cir.
2017).
Even if we agree that pCTR and pCVR are weights associated with
each of the retrieved models, the Examiner does not sufficiently explain
why, in view of Blunsom’s teaching of using scaling coefficients to avoid
floating-point underflow, it would have been obvious to generate a latent
metric model to describe user actions after a threshold amount of time “by
applying the weights to the retrieved models . . . and combining the retrieved
models after application of the weights,” as required by claim 1.
Therefore, we will reverse the rejection of claim 1. Independent
claim 12 contains similar language and for similar reasons we will also
reverse the rejection of claim 12, and of dependent claims 2–11 and 13–21.

Appeal 2020-003121
Application 15/174,865

10
CONCLUSION
The Examiner’s rejection of claims 10 and 21 under 35 U.S.C.
§ 112(b) is reversed.
The Examiner’s rejections of claims 1–21 under 35 U.S.C. § 103 are
reversed.
Specifically:
Claims
Rejected
35 U.S.C. § Reference(s)/Basis Affirmed Reversed
10, 21 112(b) Indefiniteness 10, 21
1, 2, 4–6,
10–13,
15–17, 21
103 Zhang, Blunsom 1, 2, 4–6,
10–13,
15–17, 21
3, 14 103 Zhang, Blunsom,
El-Rafei
3, 14
7–9, 18–
20
103 Zhang, Blunsom,
Karande
7–9, 18–
20
Overall
Outcome
1–21

REVERSED