Adobe Inc.

Patent Trials and Appeals Board

Aug 13, 2020

2020001643 (P.T.A.B. Aug. 13, 2020)

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.
13/746,582 01/22/2013 Marina Danilevsky 058083-0852916
(2883US01)
7838
72058 7590 08/13/2020
Kilpatrick Townsend & Stockton LLP/Adobe
Adobe Systems, Inc. 58083
Mailstop: IP Docketing - 22
1100 Peachtree Street, Suite 2800
Atlanta, GA 30309-4530
EXAMINER
DURAN, ARTHUR D
ART UNIT PAPER NUMBER
3622
NOTIFICATION DATE DELIVERY MODE
08/13/2020 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):
KTSDocketing2@kilpatrick.foundationip.com
ipefiling@kilpatricktownsend.com
PTOL-90A (Rev. 04/07)
UNITED STATES PATENT AND TRADEMARK OFFICE
____________________

BEFORE THE PATENT TRIAL AND APPEAL BOARD
____________________

Ex parte MARINA DANILEVSKY
and EUNYEE KOH
__________________

Appeal 2020-001643
Application 13/746,582
Technology Center 3600
____________________

Before JAMES P. CALVE, JEREMY M. PLENZLER, and
LEE L. STEPINA, Administrative Patent Judges.

CALVE, Administrative Patent Judge.

DECISION ON APPEAL

STATEMENT OF THE CASE
Pursuant to 35 U.S.C. § 134(a), Appellant1 appeals from the decision
of the Examiner to reject claims 20–28 and 30–41, which are all the pending
claims.2 Appeal Br. 4. We have jurisdiction under 35 U.S.C. § 6(b).
We AFFIRM.

1 “Appellant” refers to “applicant” as defined in 37 C.F.R. § 1.42. Appellant
identifies Adobe Inc. as the real party in interest. Appeal Br. 4.
2 Claims 1–19 and 29 are cancelled. See Appeal Br. 34, 37 (Claims App.).
Appeal 2020-001643
Application 13/746,582

2
CLAIMED SUBJECT MATTER
Claims 20, 27, and 34 are independent. Claim 20 is reproduced
below.
20. A method for predicting webpage features and user
features that will result in a high advertisement conversion rate,
comprising:
receiving, by an analytics engine via a network and from
a database, log data collected by a web server for a plurality of
webpages displaying same or similar advertisements, the log
data comprising webpage features and user features classified
by feature types, wherein the webpage features include data
indicating content on the webpages, and wherein the user
features include data describing user interactions with the
webpages;
generating, by the analytics engine, an information
network using the feature types of the log data, wherein the
information network comprises a star schema having a central
feature, a plurality of vertices, and a plurality of edges, wherein:
the central feature represents the same or similar
advertisements,
each of the vertices corresponds to one of the
feature types, and
each edge connects one of the vertices to the
central feature and is associated with a metric indicating
a relationship between the one of the vertices and the
central feature;
determining, by the analytics engine, a ranking of each of
the feature types corresponding to each of the vertices by
ranking the feature types in order using the metrics of the edges
connecting the vertices to the central feature;
selecting, by the analytics engine, a set of available
advertisements and using the information network as a
predictive model to determine probabilities of advertisement
conversion for the set of available advertisements, wherein the
set of available advertisements associated with the feature types
having high rankings are determined to have high probabilities
of advertisement conversion;
Appeal 2020-001643
Application 13/746,582

3
in response to receiving, via the network and by the web
server, a request from a user device to access a selected
webpage, determining, by the analytics engine, a set of features
for the selected webpage;
selecting, by the analytics engine, an available
advertisement associated with the feature types that most
closely match the set of features for the selected webpage and
that has a highest probability of advertisement conversion; and
serving, by the analytics engine, the selected available
advertisement to the web server via the network for inclusion
within the selected web page.
Appeal Br. 34–35 (Claims App.).
REJECTIONS
Claims 20–28 and 30–41 are rejected as directed to a judicial
exception to 35 U.S.C. § 101.
Claims 20–22, 24–28, and 30–41 are rejected under 35 U.S.C.
§ 103(a) as unpatentable over Agarwal (US 2010/0217648 A1, pub. Aug. 26,
2010) and Indukuri (US 2013/0163471 A1, pub. June 27, 2013).
Claim 23 is rejected under 35 U.S.C. § 103(a) as unpatentable over
Agarwal, Indukuri, and Gerace (US 5,848,396, iss. Dec. 8, 1998).
ANALYSIS
Patent Eligibility of Claims 20–28 and 30–41
Examiner’s Determination
The Examiner determines that claim limitations of receiving log data,
generating an information network, selecting an ad with a high probability of
conversion, and serving selected ad recite an abstract idea of certain methods
of organizing human activity – advertising, marketing, or sales activities or
behaviors in the independent claims. Final Act. 11; Ans. 3.
Appeal 2020-001643
Application 13/746,582

4
The Examiner determines the claims recite generic computer elements
that do not provide a practical application or a meaningful limitation to the
abstract idea but instead simply implement the abstract idea on a computer.
Final Act. 11; Ans. 3. The Examiner determines that a network, web server,
and analytics engine are generic and routine hardware structures that do not
improve technology, another technical field, or computer function, do not
use a particular machine, and do not transform or reduce a particular article
to a different state or thing. Ans. 3–4. The Examiner finds that the claimed
“feature” and “feature type” are recited so broadly as to cover any type of
feature and do not integrate the claim into a practical application. Id. at 3–5.
Appellant’s Contentions
Appellant argues that claim 20 is directed to a specific manner of data
structuring by generating an information network comprising a star schema
having a particular arrangement of a central feature, vertices, and edges that
is analogous to the self-referential table for storing and retrieving data in
Enfish, LLC v. Microsoft Corp., 822 F.3d 1327 (Fed. Cir. 2016). Appeal Br.
10–12. Appellant asserts that the claims require an information network
with a specific data structure that aggregates data to improve the accuracy of
computer selection of features of a website and solve a problem of selecting
a subset of important features from a large amount of features and build an
accurate model of predictive rules. Id. at 12.
Appellant also argues that any abstract idea recited in the claims is
integrated into a practical application by using a particular data structure of a
star schema with a central feature, vertices, and edges to determine rankings
for feature types of vertices using metrics of edges connecting vertices to
select online content (i.e., an advertisement). Id. at 13–15.
Appeal 2020-001643
Application 13/746,582

5
Principles of Law
Section 101 of the Patent Act states:
Whoever invents or discovers any new and useful
process, machine, manufacture, or composition of matter, or
any new and useful improvement thereof, may obtain a patent
therefor, subject to the conditions and requirements of this title.
35 U.S.C. § 101. This provision contains an implicit exception: “Laws of
nature, natural phenomena, and abstract ideas are not patentable.” Alice
Corp. v. CLS Bank Int’l, 573 U.S. 208, 216 (2014).
To distinguish patents that claim laws of nature, natural phenomena,
and abstract ideas from those that claim patent-eligible applications, we first
determine whether the claims are directed to a patent-ineligible concept. Id.
at 217. If they are, we consider the elements of each claim, individually and
“as an ordered combination,” to determine if additional elements “‘transform
the nature of the claim’ into a patent-eligible application” as an “inventive
concept” sufficient to ensure the claims in practice amount to significantly
more than a patent on the ineligible concept itself. See id. at 217–18.
The USPTO has issued guidance about this framework. 2019 Revised
Patent Subject Matter Eligibility Guidance, 84 Fed. Reg. 50 (Jan. 7, 2019)
(“Revised Guidance”). Under the Revised Guidance, to determine whether a
claim is “directed to” an abstract idea, we evaluate whether the claim recites:
(1) any judicial exceptions, including certain groupings of abstract ideas
listed in the Revised Guidance (i.e., mathematical concepts, certain methods
of organizing human activities such as a fundamental economic practice, or
mental processes); and (2) additional elements that integrate the judicial
exception into a practical application (see MPEP §§ 2106.05(a)–(c), (e)–(h)
(9th ed. rev. 08.2017 Jan. 2018) (“MPEP”)). Id. at 52–55.
Appeal 2020-001643
Application 13/746,582

6
Only if a claim (1) recites a judicial exception and also (2) does not
integrate that exception into a practical application, do we then consider
whether the claim (3) adds a specific limitation beyond the judicial
exception that is not “well-understood, routine, conventional” in the field
(see MPEP § 2106.05(d)) or (4) simply appends well-understood, routine,
conventional activities previously known to the industry, specified at a high
level of generality, to the judicial exception. Id. at 56.
Step 1: Is Claim 20 Within a Statutory Category?
Appellant argues the independent claims as a group. Appeal Br. 10–
19. We select claim 20 as representative. See 37 C.F.R. § 41.37(c)(1)(iv).
We address Appellant’s arguments for the dependent claims separately.
Claim 20 recites a “method” which is within a statutory category of
35 U.S.C. § 101, namely, a process. See Ans. 3. Thus, we next consider
whether claim 20 as a whole recites a judicial exception.
Step 2A, Prong One: Does Claim 20 Recite a Judicial Exception?
We agree with the Examiner that claim 20 recites an abstract idea.
The Revised Guidance enumerates this abstract idea as certain methods of
organizing human activity of commercial or legal interactions involving
advertising, marketing, or sales activities that can be performed as mental
processes––concepts performed in the human mind including observation,
evaluation, judgment, and opinion. Revised Guidance, 84 Fed. Reg. at 52.
Appellant’s disclosure addresses a business advertising problem of
matching advertisements that have the highest likelihood of users clicking
on them to webpages that users request. Spec. ¶¶ 1, 2. Toward this end, “it
can be helpful to know the website and/or user features that produce the
highest conversion (or click through) for a given advertisement.” Id. ¶ 20.
Appeal 2020-001643
Application 13/746,582

7
Thus, “the invention can be used to discover webpage features that
result in the highest rate of selection for advertisements on the webpage.”
Spec. ¶ 20. “[W]hen a user visits a webpage an advertisement that has the
highest probability for conversion can be selected from the set of advertise-
ments using the webpage and/or user features.” Id. ¶ 47. “Conversion refers
to the completion of the desired action.” Id. ¶ 24. “[T]he conversion rate for
an advertisement on a webpage may represent the percentage of the total
visitors of the webpage that click on the advertisement.” Id. The preamble
of claim 20 recites this business purpose as “[a] method for predicting
webpage features and user features that will result in a high advertisement
conversion rate.” Appeal Br. 34 (Claims App.).
The method receives data for webpages and advertisements collected
by a web server for past user activities and identifies features that resulted in
users clicking on (converting) ads. This log data of webpages and displayed
advertisements is analyzed to identify webpage and user features that led to
conversion (selection) of advertisements in the past. See id.; Spec. ¶¶ 20–24.
The first limitation of claim 20 recites this data gathering step as:
receiving, by an analytics engine via a network and from
a database, log data collected by a web server for a plurality of
webpages displaying same or similar advertisements, the log
data comprising webpage features and user features classified
by feature types, wherein the webpage features include data
indicating content on the webpages, and wherein the user
features include data describing user interactions with the
webpages.
Appeal Br. 34 (Claims App.). Such data gathering steps that just obtain
information to be analyzed recite insignificant extra-solution activity to a
judicial exception. See Revised Guidance, 84 Fed. Reg. at 55 & n.31.
Appeal 2020-001643
Application 13/746,582

8
Furthermore, such data gathering steps, when recited at a high level of
generality, can be performed as a mental process. See CyberSource Corp. v.
Retail Decisions, Inc., 654 F.3d 1366 (Fed. Cir. 2011). CyberSource held a
step of “obtaining information about other transactions that have utilized an
Internet address that is identified with the [ ] credit card transaction” “can be
performed by a human who simply reads records of Internet credit card
transactions from a preexisting database.” CyberSource, 654 F.3d at 1372;
see also Content Extraction &Transmission LLC v. Wells Fargo Bank, Nat’l
Ass’n, 776 F.3d 1343, 1347 (Fed. Cir. 2014) (holding that humans always
have performed the claimed steps of collecting data, recognizing certain data
in the set, and storing recognized data, e.g., by banks reviewing checks,
recognizing relevant data of the amount, account number, and account
holder identify, and storing the data in their records).
Here, claim 20 recites an analytics engine receiving “log data” via a
network. The log data comprises “webpage features” and “user features.”
The “webpage features include data indicating content on the webpages” and
“user features include data describing user interactions with the webpages.”
See Appeal Br. 34 (Claims App.). The “features” are recited as data types.
The Specification discloses website features as keywords, fonts, and
colors. Spec. ¶ 20. Features of website viewers are age, location, and time
of day. Id. Claim 20 recites “feature types,” which are related to webpages
and user interactions with webpages. Id. ¶ 24. “The type and/or number of
features are limitless” and “any type of feature can be used” and “can evolve
through use.” Id. ¶ 38. A person can read such features and their conversion
rates (Yes or No) for a particular advertisement (Ad Id) by observation of a
log data table such as illustrated in Appellant’s Figure 1. See id. ¶¶ 23–32.
Appeal 2020-001643
Application 13/746,582

9
“Information as such is an intangible.” Elec. Power Grp., LLC v.
Alstom S.A., 830 F.3d 1350, 1353 (Fed. Cir. 2016) (“[W]e have treated
collecting information, including when limited to particular content (which
does not change its character as information), as within the realm of abstract
ideas.”); id. at 1355 (“[M]erely selecting information, by content or source,
for collection, analysis, and display does nothing significant to differentiate
a process from ordinary mental processes, whose implicit exclusion from
§ 101 undergirds the information-based category of abstract ideas.”).
The next steps of claim 20 recite certain methods of organizing human
activity––commercial or legal interactions (including advertising, marketing,
or sales activities or behaviors). They involve the following steps:
generating, by the analytics engine, an information
network using the feature types of the log data, wherein the
information network comprises a star schema having a central
feature, a plurality of vertices, and a plurality of edges, wherein:
the central feature represents the same or similar
advertisements,
each of the vertices corresponds to one of the
feature types, and
each edge connects one of the vertices to the
central feature and is associated with a metric indicating a
relationship between the one of the vertices and the central
feature;
determining, by the analytics engine, a ranking of each of
the feature types corresponding to each of the vertices by
ranking the feature types in order using the metrics of the edges
connecting the vertices to the central feature.
Appeal Br. 34 (Claims App.).
These steps organize advertisements and features as a “star schema.”
Advertisements are arranged as a central feature. Feature types are vertices.
Their relationship to the central feature is indicated by edges. Spec. ¶ 25.
Appeal 2020-001643
Application 13/746,582

10
The star schema organizes and graphically links features. Spec. ¶ 27.
As described in the Specification, “[e]ach advertisement can be considered a
central feature, and/or a label feature in a classification problem. The links
(or edges) define a relationship between the central feature [ad] and various
features at the edges using a mathematical metric.” Spec. ¶ 25. Figure 2 of
Appellant’s disclosure is reproduced below to illustrate a star schema.

Appellant’s Figure 2 illustrates a star schema with advertisements
(AD1, AD2) as a central feature and features GEO1, GEO2, PRO1, PRO2,
URL1, URL2, SEG1, and SEG2 as vertices linked to the advertisements by
edges that define the relationship of the central feature to each feature. Spec.
¶ 25. A star schema organizes (and correlates) advertisements to associated
feature types of webpages where the ads appeared. See id. ¶¶ 20–25, 43, 47.
Appeal 2020-001643
Application 13/746,582

11
A “metric” “indicat[es] a relationship between the one of the vertices
and the central feature.” Appeal Br. 34 (Claims App.). It is described as:
A mathematical metric between at least one feature type and the
conversion of the advertisement can be calculated. . . .
. . . A plurality of edges between related feature values
within the dataset can be defined based on some metric. . . .
. . . .
. . . The links (or edges) define a relationship between
the central feature and various features at the edges using a
mathematical metric. . . .
The term “mathematical metric” is used herein to refer to
anything that defines a mathematical relationship between a
distinct pair of feature types. The links between features and
advertisements can be a mathematical metric that relates the
feature with the advertisement. For example, the mathematical
metric can be the co-occurrence, correlation, Euclid[i]an
distance, or cosine similarity of the feature and the conversion
of the advertisement as described in more detail below.
Spec. ¶¶ 4, 5, 25, 26. A mathematical metric is described, but claim 20 only
recites “a metric” indicating relationships of vertices and a central feature.
Claim 20 recites the analytics engine ranking each feature type using
the metrics of edges without reciting how this is done. The Specification
indicates only that an edge can be the “lift” of a feature on the conversion
rate with few other details, and other metrics may be used like co-occurrence
or cosine similarity. Id. ¶ 52. “[F]eatures can be ranked based on feature
type and features related to a high conversion can be returned.” Id. ¶ 27.
Recited at such a high level of generality as in claim 20, the steps can
be performed as a mental process. See Elec. Power, 830 F.3d at 1354
(“[W]e have treated analyzing information by steps people go through in
their minds, or by mathematical algorithms, without more, as essentially
mental processes within the abstract-idea category.”).
Appeal 2020-001643
Application 13/746,582

12
“Furthermore, with the exception of generic computer-implemented
steps, there is nothing in the claims themselves that foreclose them from
being performed by a human, mentally or with pen and paper.” Intellectual
Ventures I LLC v. Symantec Corp., 838 F.3d 1307, 1318 (Fed. Cir. 2016).
Here, a person can organize advertisements and feature types in a star
schema topology as a mental process or using pen and paper. A person can
determine a metric of a “relationship” between vertices and a central feature
advertisement, e.g., whether a feature was used with an advertisement that
was converted in the historical log data. See Spec. ¶¶ 20–25. A person can
rank features that provide the highest selection rate for advertisements on a
webpage by observing advertisements that were clicked on more often by
users than other ads. Id. ¶¶ 24–27. Appellant’s Figure 1 illustrates log data
in tables with feature types correlated to ads and conversions. Id. ¶¶ 23–32.
In CyberSource, claimed steps of obtaining information about credit
card transactions could be performed by a person simply reading records of
internet transactions and constructing a map of credit card numbers by
writing them down from a particular Internet address. CyberSource, 654
F.3d at 1372. A person also could observe that different transactions made
using different credit cards with different names and billing addresses all
originated from the same Internet address. Id.
Here, a person can observe webpage features that were used with each
advertisement and identify features associated with conversions of ads in the
log data. Spec. ¶¶ 23, 24, Fig. 1. A person can organize advertising activity
data using pen and paper to produce a star schema like Appellant’s Figure 2
above. Essentially, a star schema indexes feature types and advertisements
into a data structure with conversion results. Id. ¶ 27.
Appeal 2020-001643
Application 13/746,582

13
In an analogous context that used HTML tags and metafiles to create
an index of records in a database, such steps of data organization were not
patent-eligible when recited at such a high level of generality.
As the patent itself observes, the invention relates to “locating
information in a database, and . . . using an index that includes
tags and metafiles to locate the desired information.” Id. at col.
1 ll. 24–26. This type of activity, i.e., organizing and accessing
records through the creation of an index-searchable database,
includes longstanding conduct that existed well before the
advent of computers and the Internet. For example, a hardcopy-
based classification system (such as library-indexing system)
employs a similar concept as the one recited by the ’434 patent.
There, classifiers organize and cross-reference information and
resources (such as books, magazines, or the like) by certain
identifiable tags, e.g., title, author, subject. Here, tags are
similarly used to identify, organize, and locate the desired
resource.
Intellectual Ventures I LLC v. Erie Indem. Co., 850 F.3d 1315, 1327 (Fed.
Cir. 2017) (emphasis added).
The step of “ranking the feature types in order using the metrics of the
edges” is recited so broadly that a person can perform the step as a mental
process or with pen and paper. See Elec. Power, 830 F.3d at 1354; see also
Braemar Mfg., LLC v. The ScottCare Corp., Appeal No. 2019-2263, 2020
WL 3564687, at *4 (Fed. Cir. July 1, 2020) (determining a “measure of
merit” of a cardiac condition by, at most, executing a mathematical formula
or selecting a value from a lookup table recites a mental process).
These steps filter and classify feature types according to whether the
feature types appear with advertisements that are converted by a user of the
webpage (i.e., clicked-on or selected) and how frequently such conversions
occurred for particular feature types and advertisements.
Appeal 2020-001643
Application 13/746,582

14
Claim 20 thus collects log data for webpages and advertisements and
organizes/classifies the data into a star schema that is used to determine the
probability of advertisements being converted when used with a feature type.
Similar claims were patent-ineligible in TLI Communications. See In re TLI
Commc’ns LLC Patent Litig., 823 F.3d 607 (Fed. Cir. 2016). It was held:
[T]he claims, as noted, are simply directed to the abstract idea
of classifying and storing digital images in an organized manner.
Consistent with the Supreme Court’s rejection of “categorical
rules” to decide subject matter eligibility, . . . we have applied
the “abstract idea” exception to encompass inventions pertaining
to methods of organizing human activity. . . . Here, we find that,
like the claims at issue in Content Extraction which were
directed to “collecting data,” “recognizing certain data within
the collected data set,” and “storing the recognized data in
memory,” 776 F.3d at 1347, attaching classification data, such
as dates and times, to images for the purpose of storing those
images in an organized manner is a well-established “basic
concept” sufficient to fall under Alice step 1.
TLI, 823 F.3d at 613 (emphasis added); see Berkheimer v. HP Inc., 881 F.3d
1360, 1367 (Fed. Cir. 2018) (“The parsing and comparing of claims 1–3 and
9 are similar to the collecting and recognizing of Content Extraction . . . ,
and the classifying in an organized manner of TLI, 823 F.3d at 613.”).
Here, log data is filtered to extract features and advertisements that are
organized into a star schema that is compared to “available advertisements”
and their filtered features to select ads whose features match closest to the
features of a requested webpage with the highest conversion. See BASCOM
Glob. Internet Servs., Inc. v. AT&T Mobility LLC, 827 F.3d 1341, 1348
(Fed. Cir. 2016) (“We agree with the district court that filtering content is an
abstract idea because it is a longstanding, well-known method of organizing
human behavior, similar to concepts previously found to be abstract.”).
Appeal 2020-001643
Application 13/746,582

15
The Specification describes filtering used to select feature types. See
Spec. ¶¶ 7, 48, 60. Claim 20 recites this process as selecting feature types
and ranking feature types by metrics of their relationship to advertisements
in a central feature. As a result of this filtering, more highly ranked features
correlate with a greater likelihood of advertisement conversion. See id. ¶ 63.
The final steps of claim 20 employ the star schema as a “predictive
model to determine probabilities of advertisement conversion for the set of
available advertisements.” Appeal Br. 34 (Claims App.). Advertisements
have a higher probability of conversion if they are associated with feature
types of the star schema that historically provided higher probabilities of
advertisement conversion. The star schema organizes and ranks webpage
feature types by probability of conversion so new available advertisements
are provided with user-selected webpages whose features match features
associated with high conversion rates of such advertisements in the past. No
other details are recited. Past conversions rates of ads used with certain
features is used to predict future conversion rates of similar ads and features.
In the final steps, “a request from a user device to access a selected
webpage” is received, and “a set of features for the selected webpage” is
determined so “an available advertisement associated with the feature types
that most closely match the set of features for the selected webpage and that
has a highest probability of advertisement conversion” can be selected and
served to the web server via the network for inclusion within the selected
web page. Id. at 34–35. These steps can be performed as mental processes
of observation, evaluation, and judgment. A person can match features of a
selected webpage to advertisements that produced the highest conversion in
the past with such webpage features using the star schema index structure.
Appeal 2020-001643
Application 13/746,582

16
“The predictive model can assume, for example, that if a user is
served an advertisement associated with such features they will be more
likely to click on the advertisement if the user and/or webpage exhibit more
of the highly ranked features for this advertisement.” Spec. ¶ 46. These
steps also involve mental processes. According to the Specification:
[A] predictive model can be developed that associates each
advertisement in a set of advertisements with features that
produce a high probability of advertisement conversion. That
is, a circumstance can be returned that specify the features (or
feature types or informative feature types) that produce a high
probability of advertisement conversion. Then, when a user
visits a webpage an advertisement that has the highest
probability for conversion can be selected from the set of
advertisements using the webpage and/or user features.
Spec. ¶ 62. This description of the predictive model indicates the analysis,
filtering, and matching can be performed as a mental process by observation,
evaluation, judgment, and opinion. Revised Guidance, 84 Fed. Reg. at 52.
Essentially, the method of claim 20 optimizes the conversion rates of
advertisements placed on webpages requested by users by collecting and
analyzing historical log data to identify feature types associated with higher
rates of conversion of particular advertisements. The historical log data is
used like survey data of past users’ conversions rates of advertisements on
webpages with features types. As the Specification describes this process:
[T]he predictive model can create rules for future ad serving
needs . . . [and] can help marketers generate more conversion
for their ads by providing information to more strategically
place ads and/or can create more revenue for webpage owners
by selling advertisement space with a higher likelihood of
conversion.
Id. ¶ 47. In other words, the predictive model provides a business solution.
Appeal 2020-001643
Application 13/746,582

17
In OIP Technologies, similar claims to offer-based price optimization
were held to be patent-ineligible. OIP Techs., Inc. v. Amazon.com, Inc., 788
F.3d 1359, 1361–62 (Fed. Cir. 2015). The method involved steps of testing
a plurality of prices for products by sending offers to potential customers,
gathering statistical data of how customers reacted to the offers testing the
prices, using the data to estimate outcomes by mapping a demand curve over
time for a product, and automatically selecting and offering a new price
based on the estimated outcome. Id. This method of organizing human
activity was similar to fundamental economic concepts held to be abstract
ideas. Id.; Revised Guidance, 84 Fed. Reg. at 52 n.13 (citing OIP).
Here, claim 20 extracts feature types, advertisements, and conversion
rates from log data and analyzes the data to identify feature types associated
with the highest conversion rates for particular advertisements. Then, future
advertisements are served with those webpages whose features provide the
highest probability of conversion for that advertisement based on historical
log data. See BSG Tech LLC v. BuySeasons, Inc., 899 F.3d 1281, 1286 (Fed.
Cir. 2018) (users considering historical usage data when entering data into a
database is a patent-ineligible method of organizing activity).
Furthermore, “embodiments of the invention are directed toward
feature selection in a network framework [but] [w]hile the example of
determining advertisement conversion probabilities based on webpage
features is used throughout this disclosure, embodiments of the invention
may be used in many other applications.” Spec. ¶ 28.
Accordingly, we determine that claim 20 recites an abstract idea of
certain methods of organizing human activity as commercial interactions for
advertising, marketing, and sales activities and mental processes.
Appeal 2020-001643
Application 13/746,582

18
Dependent Claims
We agree with the Examiner that dependent claims 21–26, 28, 30–33,
and 35–41 recite other details of this judicial exception. Final Act. 21. For
example, claim 21 recites edge metrics “based on a conversion rate of each
of the same or similar advertisements represented by the central feature.”
Claim 22 recites the edge metric is a co-occurrence, correlation, Euclidian
distance, or cosine similarity. Claim 23 recites weights for some edges and
the weights indicate a “confidence of the edge’s metric.” Claim 24 recites
“multiple central features belonging to multiple classes.” Claim 25 recites
the probabilities of advertisement conversion are determined for each class.
Claim 26 determines probabilities by altering weights. See Ans. 5.
It is well-settled that “[a]n abstract idea can generally be described at
different levels of abstraction.” Apple, Inc. v. Ameranth, Inc., 842 F.3d
1229, 1240 (Fed. Cir. 2016). Reciting other abstract features of the abstract
idea recited in claim 20 in dependent claims does not make the dependent
claims patent-eligible. The dependent claims recite other abstract features of
the abstract idea recited in claim 20 and the other independent claims.
Given such a high-level recital of the dependent claim features, the
Examiner’s determination that the dependent claims recite other features of
the identified abstract idea is supported by a preponderance of evidence. We
have discussed the abstract nature of the metric and classification/filtering
features. The weights represent a trustworthiness or confidence of links and
can be specified with user guidance or learned by optimization. Spec. ¶ 53.
We addressed the abstractness of optimization in our discussion of claim 20.
Accordingly, we determine that the dependent claims recite other
aspects of the judicial exception recited in claim 20.
Appeal 2020-001643
Application 13/746,582

19
Step 2A, Prong Two: Integration into a Practical Application
We next determine whether claim 20 recites additional elements that
integrate the abstract idea into a practical application. Revised Guidance, 84
Fed. Reg. at 54 (Revised Step 2A, Prong Two). We determine that claim 20
lacks additional elements that improve a computer or other technology. The
additional elements do not implement the abstract idea in conjunction with a
particular machine or manufacture that is integral to the claim. They do not
transform or reduce a particular article to a different state or thing. They do
not apply the abstract idea in a meaningful way beyond merely linking it to a
particular technological environment. See Revised Guidance, 84 Fed. Reg.
at 55 and MPEP sections cited therein.
We recognize that “[s]oftware can make non-abstract improvements
to computer technology just as hardware improvements can, and sometimes
the improvements can be accomplished through either route.” Enfish, LLC
v. Microsoft Corp., 822 F.3d 1327, 1335 (Fed. Cir. 2016); see Appeal Br. 11.
However, in this regard, “to be directed to a patent-eligible improvement to
computer functionality, the claims must be directed to an improvement to
the functionality of the computer or network platform itself.” Customedia
Techs., LLC v. Dish Network Corp., 951 F.3d 1359, 1365 (Fed. Cir. 2020)
(citing Enfish, 822 F.3d at 1336–39). The holding in Enfish illustrates why
claim 20 here is not patent-eligible.
In Enfish, the claims recited a self-referential database that stored all
entity types in a single table and defined the table’s columns by rows in the
same table. Enfish, 822 F.3d at 1332. The self-referential database allowed
faster searching of data than a relational model and allowed more effective
data storage including unstructured text and images. Id. at 1333.
Appeal 2020-001643
Application 13/746,582

20
The self-referential database model also allowed more flexibility in
configuring a database than a relational database. Id. It could be launched
with little-or-no column definitions because columns could be added simply
by inserting a new row that instigates creation of a new column. Id.
The self-referential database accomplished these improvements with a
new database structure based on a “means for configuring” algorithm that
assigned each row and column an object identification number that acted as
a pointer to the associated row or column and stored information about a
column in one or more rows “rendering the table self-referential” so new
columns were available for immediate use through creation of new column
definition records in a row(s). Id. at 1336–37. The self-referential database
thus provided increased flexibility, faster search times, and smaller memory
requirements compared to conventional databases. Id. at 1337. Because the
algorithm stored information related to a column in a row of the same table,
“new columns can be added by creating new rows in the table.” Id. at 1338.
The claims were patent-eligible because “the self-referential table recited in
the claims on appeal is a specific type of data structure designed to improve
the way a computer stores and retrieves data in memory.” Id. at 1339.
Here, the star schema organizes data of feature types, advertisements,
and their relationships as a data structure index without improving computer
or database operations or data retrieval. Appellant does not assert to have
invented a star schema data structure. Nor does Appellant assert that a star
schema enables computers or networks to operate faster or more efficiently,
or that it stores data more efficiently or accessibly. We have no evidence
that a star schema organizes or analyzes data any better than a conventional
database or that it identifies data relationships between data items better.
Appeal 2020-001643
Application 13/746,582

21
The features identified by Appellant as an improved data structure,
namely, the central feature, vertices, and edges, are recited at a high level as
a judicial exception. “It has been clear since Alice that a claimed invention’s
use of the ineligible concept to which it is directed cannot supply the
inventive concept that renders the invention ‘significantly more’ than that
ineligible concept.” BSG Tech, 899 F.3d at 1290; see id. at 1291 (“As a
matter of law, narrowing or reformulating an abstract idea does not add
‘significantly more’ to it.”); see RecogniCorp, LLC v. Nintendo Co., 855
F.3d 1322, 1327 (Fed. Cir. 2017) (“Adding one abstract idea (math) to
another abstract idea (encoding and decoding) does not render the claim
non-abstract.”); Synopsys, Inc. v. Mentor Graphics Corp., 839 F.3d 1138,
1151 (Fed. Cir. 2016) (“But, a claim for a new abstract idea is still an
abstract idea.”); Versata Dev. Grp., Inc. v. SAP Am., Inc., 793 F.3d 1306,
1335 (Fed. Cir. 2015) (holding claims that improved an abstract idea but did
not recite the supposed computer improvements were not patent eligible);
Revised Guidance, 84 Fed. Reg. at 55 n.24 (additional elements refer to
claim features, limitations, and/or steps that are recited in a claim beyond the
identified judicial exception).
Even if we consider these features to be “additional elements,” we are
unpersuaded that they improve the function of computers or technology, that
they are used with a particular machine integral to the claim, or that they
transform a particular article to a different state or thing. Appellant argues
that aggregating data with this data structure improves the accuracy of
computer selection of features of a website and “solves the problem of
selecting a subset of important features from a potentially large amount of
features in order to build an accurate model.” Appeal Br. 12.
Appeal 2020-001643
Application 13/746,582

22
These arguments are not commensurate with the scope of claim 20,
which does not require potentially large amounts of features. Instead, it
recites “log data” “for a plurality of webpages displaying same or similar
advertisements” where the log data comprises “webpage features” and “user
features.” Appeal Br. 34 (Claims App.). A plurality of webpages can be
two or more webpages as can advertisements and webpage features.
Nor does the Specification indicate that the star schema improves data
processing or computer or database functions. The Specification indicates
that “[v]arious other datasets can be used” and “may vary depending on the
type of web server, client needs, the type of advertisement, etc.” Spec. ¶ 33.
Other features may be relevant to ad conversion including nominal features,
continuous features, client device features, domain specific features, prefix
features, geolocation features, environmental features, profile script features,
segment features, URL features, referral features, and time features. Id.
¶¶ 34–37. However, claim 20 recites only webpage and user features. Thus,
claim 20 does not collect or analyze a large number of different features that
improve the predictability of advertisement conversion. Any feature can be
used such as client defined phrase usage, word usage, keyword information,
type of sport, hobby type, or subject matter (id. ¶ 38), but claim 20 organizes
and analyzes only two types of features––webpage and user features.
Claim 20 recites a solution to a business problem not to a technical
problem as in Enfish. The business problem is matching ads to webpages to
optimize conversion rates. The claimed solution matches ads to webpages
where similar ads were converted in the past without claiming any details of
how ads and features are matched using technical improvements beyond the
star schema that indexes and ranks ads and features as a judicial exception.
Appeal 2020-001643
Application 13/746,582

23
The star schema provides a generic format to organize historical data
for advertisements and feature types with “a relationship” between features
and advertisements. A “metric indicating a relationship between the one of
the vertices and the central feature” does not improve computer or database
operations. Ranking feature types using metrics of edges is not a technical
improvement when recited at a high level of generality without any technical
details. A predictive model that determines probabilities of advertisement
conversion merely matches new advertisements to webpage features that led
to conversion of similar ads in the past without reciting any technical details.
Even if the Specification describes improvements to computers or
databases, these features are not recited in claim 20. See ChargePoint, Inc.
v. SemaConnect, Inc., 920 F.3d 759, 769–70 (Fed. Cir. 2019); Ericsson, 955
F.3d 1317, 1325 (Fed. Cir. 2020) (holding that the specification must yield
to the claim language when identifying the true focus of a claim); Synopsys,
839 F.3d at 1149 (“The § 101 inquiry must focus on the language of the
Asserted Claims themselves.”); Accenture Glob. Servs., GmbH v. Guidewire
Software, Inc., 728 F.3d 1336, 1345 (Fed. Cir. 2013) (“[T]he important
inquiry for a § 101 analysis is to look to the claim.”). Instead, claim 20
recites method steps at a high level of generality as a judicial exception.
“[T]he fact that the required calculations [may] be performed more
efficiently via a computer does not materially alter the patent eligibility of
the claimed subject matter.” Bancorp Servs., L.L.C. v. Sun Life Assurance
Co. of Can. (U.S.), 687 F.3d 1266, 1277–78 (Fed. Cir. 2012) (holding that
the performance by a computer of operations that previously were performed
manually or mentally, albeit less efficiently, does not convert an abstract
idea into eligible subject matter).
Appeal 2020-001643
Application 13/746,582

24
“[T]he mere recitation of a generic computer cannot transform a
patent-ineligible abstract idea into a patent-eligible invention.” Alice, 573
U.S. at 223; see also DDR Holdings, LLC v. Hotels.com, 773 F.3d 1245,
1256 (Fed. Cir. 2014)(“[T]hese claims [of prior cases] in substance were
directed to nothing more than the performance of an abstract business
practice on the Internet or using a conventional computer. Such claims are
not patent-eligible.”); buySAFE v. Google, Inc., 765 F.3d 1350, 1355 (Fed.
Cir. 2014) (“That a computer receives and sends the information over a
network—with no further specification—is not even arguably inventive.”).
The claimed “analytics engine,” “network,” “database,” “web server,”
and computer system are described as generic components used to perform
generic functions. Analytics engine 310 can be used to analyze data stored
in database 308. Spec. ¶ 42. Analytics engine 310 can execute different
processes and algorithms when analyzing the data. Id. The network system
can be used to determine webpage features that are related to ads that have a
high conversion probability. Id. ¶ 39. Web server 306 hosts a webpage. Id.
It can be a single server, plurality of servers, plurality of distributed servers,
and/or a plurality of servers spread across a network cloud, and one or more
webpages can be hosted on webserver 306. Id. Webserver 306 also can be
coupled communicatively with database 308. Id. ¶ 41. Database 308 can
store log data collected at webserver 306. Id., Fig. 1. These elements do not
provide a particular machine that is integral to the claim or transform an
article to a different state or thing. See Intellectual Ventures I LLC v.
Capital One Bank (USA), 792 F.3d 1363, 1366 (Fed. Cir. 2015) (“An
abstract idea does not become nonabstract by limiting the invention to a
particular field of use or technological environment, such as the Internet.”).
Appeal 2020-001643
Application 13/746,582

25
Essentially, claim 20 falls within a familiar class of claims that recite
a result without reciting details of how the result is accomplished through
advances in computers, databased, networks, or other technology. Claim 20
recites receiving log data (information as such), organizing the data as a star
schema with a metric used to rank a relationship of feature types, and using
the information network as a predictive model to determine probabilities of
advertisement conversion. Notably absent from claim 20 is a recital of any
hardware or software improvements used to form a star schema, calculate a
metric, rank feature types using the metrics of edges, or use the information
network (star schema) as a predictive model. If the Specification describes
technical improvements for these elements, they are not recited in claim 20.
Therefore, they cannot be relied on to integrate the judicial exception.
Nor does the Specification establish that the selection of targeted ads
is improved. See FairWarning IP, LLC v. Iatric Sys., Inc., 839 F.3d 1089,
1095 (Fed. Cir. 2016) (“While the claimed system and method certainly
purport to accelerate the process of analyzing audit log data, the speed
increase comes from the capabilities of a general-purpose computer, rather
than the patented method itself.”); OIP, 788 F.3d at 1363 (“But relying on a
computer to perform routine tasks more quickly or more accurately is
insufficient to render a claim patent eligible.”); Intellectual Ventures I LLC
v. Erie Indem. Co., 711 F. App’x 1012, 1017 (Fed. Cir. 2017) (“Though the
claims purport to accelerate the process of finding errant files and to reduce
error, we have held that speed and accuracy increases stemming from the
ordinary capabilities of a general-purpose computer ‘do[ ] not materially
alter the patent eligibility of the claimed subject matter.”’).
Appeal 2020-001643
Application 13/746,582

26
In BSG, considering historical usage information while inputting data
into a database to ensure consistency recited an abstract idea of organizing
human activity. See BSG, 899 F.3d at 1286 (“It amounts to having users
consider previous item descriptions before they describe items to achieve
more consistent item descriptions. Whether labeled as a fundamental, long-
prevalent practice or a well-established method of organizing activity, this
qualifies as an abstract idea.”). Even “the recitation of a database structure
slightly more detailed than a generic database [did] not save the asserted
claims at step one [of Alice].” Id. at 1287. This holding is pertinent to
Appellant’s arguments that the star schema data structure is patent-eligible.
The patent in BSG claimed “summary comparison usage information”
broadly. Id. The patent specification stated that “‘usage’ is employed herein
in its broadest possible sense to include information relating to occurrence,
absolute or relative frequency, or any other data which indicates the extent
of past usage with respect to the various choices.” Id. Here, claim 20 recites
“edge,” “metric,” and “ranking” based on usage of features. These elements
are described very broadly in Specification as concepts that include any type
of feature and many metrics and rankings. See Spec. ¶¶ 4, 5, 24–27, 52, 63.
It is well-settled that “a claim is not patent eligible merely because it
applies an abstract idea in a narrow way.” BSG, 899 F.3d at 1287. Any
improvements that may result from using historical summary comparison
information to guide user selection of classifications, parameters, and values
or increase consistency, speed, and efficiency in accessing many records to
find only the most relevant few records were “not improvements to database
functionality.” Id. at 1288. “[T]hey are benefits that flow from performing
an abstract idea in conjunction with a well-known database structure.” Id.
Appeal 2020-001643
Application 13/746,582

27
Here, the star schema, as claimed, provides a generic data structure to
organize and store advertising and ecommerce data according to pre-existing
relationships. The edges and metrics record pre-existing relationships of the
feature types and advertisements derived from the log data without technical
details of how that process occurs. Nor does claim 20 recite technical details
of how the star schema facilitates efficient data processing and selection of
new advertisements with the highest probability of conversion.
Appellant argues that the last three limitations of determining a set of
features for a webpage in response to receiving an access request, selecting
an advertisement associated with feature types that most closely match the
features of the webpage, and serving the advertisement to a web server for
inclusion in the web page integrate the abstract idea by selecting and serving
customized web content, which is a computer-specific function rather than
an abstract idea. Appeal Br. 14.
Providing customized information as targeted advertising based on
user or webpage features does not make a judicial exception patent-eligible.
See Affinity Labs of Tex., LLC v. Amazon.com Inc., 838 F.3d 1266, 1271
(Fed. Cir. 2016) (providing targeted advertising based on user demographic
information without any solution to a technological problem is not patent-
eligible); Intellectual Ventures I LLC v. Capital One Bank (USA), 792 F.3d
1363, 1369–70 (Fed. Cir. 2015) (tailoring user content based on information
known about a user such as location and specific data is another fundamental
practice long prevalent in our society even when performed on the Internet).
In Bridge and Post, a method of serving tailored advertising based on
a unique device identifier of a user and derived geographic location and
demographic information did not make a judicial exception patent eligible.
Appeal 2020-001643
Application 13/746,582

28
Bridge and Post does not claim to have invented new
networking hardware or software. Nor does it claim to have
invented HTTP header fields, user identifiers, encryption
techniques, or any other improvement in the network
technology underlying its claims. The specification states that
the invention filled a need for a system which would “ensure
higher access rates, longer browse times, and increased
consumption of media” by users. . . . But each of these goals is
in the abstract realm—an improvement in the success or
monetization of tracking users with personalized markings in
order to serve advertisements—not an improvement in
networking or computer functionality. None of these alleged
improvements “enables a computer . . . to do things it could not
do before.” Finjan, Inc. v. Blue Coat Sys., Inc., 879 F.3d 1299,
1305 (Fed. Cir. 2018) (emphasis added). Such claims, whose
focus is “not a physical-realm improvement but an
improvement in a wholly abstract idea,” are not eligible for
patenting. SAP Am., Inc. v. InvestPic, LLC, 898 F.3d 1161,
1168 (Fed. Cir. 2018).
Bridge & Post, Inc. v. Verizon Commc’ns, Inc., 778 F. App’x 882, 889 (Fed.
Cir. 2019). Appellant argues that “[t]he Specification further explains that a
‘feature dataset can be collected and transformed into an information
network . . . that respects feature type differences yet maintains correlation
between features . . . allowing for both the identification of valuable
features, and/or the ability to construct predictive rules for future data
records.’” Appeal Br. 12. However, claim 20 does not recite any technical
improvement used to accomplish this result, e.g., by making computers run
faster or more efficiently or identifying and retrieving data more efficiently.
Accordingly, we determine that claim 20 does not include additional
elements that integrate the judicial exception into a practical application.
Appeal 2020-001643
Application 13/746,582

29
Step 2B: Does Claim 20 Include an Inventive Concept?
We next consider if claim 20 recites additional elements, individually,
or as an ordered combination, that provide an inventive concept. Alice, 573
U.S. at 217–18. The second step of the Alice test is satisfied when the claim
limitations involve more than performance of well-understood, routine, and
conventional activities previously known to the industry. Berkheimer v. HP
Inc., 881 F.3d at 1367; see Revised Guidance, 84 Fed. Reg. 56 (explaining
that the second step of the Alice analysis considers whether a claim adds a
specific limitation beyond a judicial exception that is not “well-understood,
routine, conventional” activity in the field).
Individually, the additional elements perform the abstract idea using
generic components to perform generic functions of receiving and analyzing
data at a high level of generality. Even if these steps are groundbreaking,
innovative, or brilliant, that is not enough for eligibility. See Ass’n for
Molecular Pathology v. Myriad Genetics, Inc., 569 U.S. 576, 591 (2013);
accord SAP Am., 898 F.3d at 1163 (“No matter how much of an advance in
the finance field the claims recite, the advance lies entirely in the realm of
abstract ideas, with no plausibly alleged innovation in the non-abstract
application realm. An advance of that nature is ineligible for patenting.”).
As an ordered combination, the additional elements provide no more
than when they are considered individually. Alice, 573 U.S. at 225. They
recite generic computer components that perform generic functions. They
are used as tools to implement the judicial exception. SAP Am., 898 F.3d at
1169–70 (holding limitations of various databases and processors did not
improve computers but used already available computers and available
functions as tools to execute the claimed process); see Final Act. 11.
Appeal 2020-001643
Application 13/746,582

30
“The ‘novelty’ of any element or steps in a process, or even of the
process itself, is of no relevance in determining whether the subject matter of
a claim falls within the § 101 categories of possibly patentable subject
matter.” Diamond v. Diehr, 450 U.S. 175, 188–89, (1981). “An abstract
idea can generally be described at different levels of abstraction.” Apple,
842 F.3d at 1240; see also Western Express Bancshares v. Green Dot Corp.,
Appeal No. 2020-1079, 2020 WL 3967855, *3 (Fed. Cir. July 14, 2020)
(“But the absence of the exact invention in the prior art does not prove the
existence of an inventive concept.”).
The holding in Electric Power Group applies here. “Nothing in the
claims, understood in light of the specification, requires anything other than
off-the-shelf, conventional computer, network, and display technology for
gathering, sending, and presenting the desired information.” Elec. Power,
830 F.3d at 1355; see id. at 1354 (“The claims . . . fit into the familiar class
of claims that do not ‘focus . . . on [] an improvement in computers as tools,
but on certain independently abstract ideas that use computers as tools.’”).
As discussed under Prong Two, the Specification describes the analytics
engine, network, database, log data, predictive model, and other components
in a generic fashion indicating that they provide no advance to computers or
other technology. Nor has Appellant asserted that the predictive model is
innovative beyond its use as a tool to perform steps of the judicial exception.
The dependent claim features suffer from the same infirmity as claim 20.
Accordingly, we determine that claim 20 lacks an inventive concept
sufficient to transform the abstract idea into patent eligible subject matter.
Thus, we sustain the rejection of claims 20–28 and 30–41 as directed to a
judicial exception under 35 U.S.C. § 101.
Appeal 2020-001643
Application 13/746,582

31
Claims 20–22, 24–28, and 30–41
Rejected over Agarwal and Indukuri
Appellant argues the rejection of the claims as a group except for
dependent claims 21, 24, and 26. See Appeal Br. 21–32. We select claim 20
as representative of the group (see 37 C.F.R. § 41.37(c)(1)(iv)) and address
Appellant’s arguments for claims 21, 24, and 26 separately.
Regarding claim 20, the Examiner finds that Agarwal discloses the
claimed method of receiving log data in the form of webpage features such
as keywords, subjects/categories, and content of webpages and user features
such as user interactions with web pages/web advertisements to determine a
probability a user will click on a text web advertisement. Final Act. 13–14.
The Examiner also finds that these features are correlated to a central feature
of a text web advertisement to determine a metric/weight of each particular
feature to a particular ad that indicates a relationship between the feature and
ad such as a probability that a user will click on the text web advertisement.
Id. at 13–15. The Examiner reasons that Agarwal’s score/metric between a
webpage feature type and an ad functions as “the edge” because it describes
a score/number for a conversion rate of an ad based on the feature. Ans. 6.
Appellant argues that Agarwal discloses expert statistical models that
can be generated based on a linear model family and therefore teaches away
from a star topology. Appeal Br. 22–23. Appellant argues that Agarwal and
Indukuri fail to teach or suggest “edges connecting vertices to the central
feature” because Agarwal discloses relevance scores for web pages and their
associated advertisements are a feature of a web page. Id. at 23. Appellant
also argues that Agarwal does not determine rankings of webpage features
using scores between webpage feature types and ads. Id. at 24–25.
Appeal 2020-001643
Application 13/746,582

32
We do not agree that Agarwal’s teaching that expert statistical models
can be generated based on a logistic regression model or generalized linear
model family teaches away from a star topology recited in claim 20. Prior
art teaches away when a skilled artisan reading the reference would be
discouraged from following the path set out in the reference or would be led
in a direction divergent from the path taken by an applicant. In re Gurley,
27 F.3d 551, 553 (Fed. Cir. 1994). A lack of discussion of a feature does not
mean a reference teaches away from it. DyStar Textilfarben GmbH & Co.
Deutschland KG v. C.H. Patrick Co., 464 F.3d 1356, 1364 (Fed. Cir. 2006).
Here, the claimed star schema organizes activities for advertising and
marketing by representing advertisements as a central feature, webpage/user
features as vertices, and edges connecting each vertex to the central feature
as a metric indicating the relationship between each vertex and the central ad
feature. See Appeal Br. 34 (Claims App.); Spec. ¶ 25. The metric defines a
relationship between the webpage feature and an advertisement such as a co-
occurrence, correlation, Euclidian distance, or cosine similarity. Spec. ¶ 26.
The “star schema” includes a central feature advertisement tied to individual
feature types at vertices. Each vertex is connected to the central feature by a
single edge that indicates a relationship between that vertex and the central
advertisement feature.
Agarwal also scores the relationship of webpages and advertisements
to determine how closely a web page matches an advertisement. Agarwal
¶¶ 30–35. Agarwal scores a webpage and an advertisement along multiple
categories or criteria to determine their relevance/similarity along multiple
semantic axes measurable by a cosine similarities. See id. The axes thus
correspond to the claimed edges that connect the vertices and central ad.
Appeal 2020-001643
Application 13/746,582

33
Agarwal also recognizes that webpages and advertisements comprise
multiple features that may be related to one another with different degrees of
relevance. Agarwal scores the relevance of web pages and advertisements to
one another and analyzes each semantic value/axis using cosine similarities
to determine a similarity of a webpage and its features to an advertisement.
Id. Then, Agarwal weights the features to predict the clickability of a web
advertisement on a webpage with those features. Id. ¶ 36.
Agarwal determines a probability that a user will click on a text web
advertisement on a particular web page based on the features of the webpage
and advertisement so web advertisements are allocated to appropriate web
pages to increase revenues paid to a provider of the web advertisements if a
user clicks on the web advertisement. Agarwal ¶ 15. Agarwal analyzes user
features from a log of historical user interactions with web pages and web
advertisements. Id. ¶ 40. User interaction features include using a computer
mouse, microphone, keyboard, trackball, or other interface. Id. ¶ 15; Spec.
¶ 38.
Agarwal obtains webpage features such as keywords, their frequency
of use on a webpage, a subject matter category of a webpage, keywords in a
title of a webpage, keyword phrases, and publisher identifier. Id. ¶¶ 16, 25–
29. Agarwal scores each feature compared to a central advertisement using
metrics of a cosine similarity or semantic value. Id. ¶¶ 30–35. A cosine
similarity of 1.0 indicates a high relevance between certain keyword features
of a webpage and an advertisement so the advertisement may be determined
to be relevant. Id. ¶¶ 30, 31. The similarity of keywords in a webpage and
advertisement may be expressed as a semantic value and other semantic
values may be used for keyword phrase features. Id. ¶¶ 31–33.
Appeal 2020-001643
Application 13/746,582

34
We agree with the Examiner that the cosine similarity scores/semantic
values correspond to the claimed edge metrics that indicate a relationship
between a webpage feature (vertex) and an advertisement (central feature).
Ans. 6–9. These “relevance scores” can be used to predict the clickability of
a web advertisement placed on the web page. Agarwal ¶ 36. Agarwal
provides expert statistical models “adapted to determine a click-through-rate
for a web page based on weightings assigned to various features of the web
page.” Id. ¶ 37. The models are “generated to explain interaction between
features on web pages and features on web advertisements [and] may be
utilized to predict a probability of whether a user is going to click on a web
advertisement on a web page.” Id. ¶ 41. Agarwal uses models as metrics to
correlate and rank the relationship of webpage features to an advertisement
just like the claimed “edges.” The metric may be a cosine similarity or co-
occurrence of keywords in a webpage. Id. ¶ 30; see Spec. ¶ 50 (metrics are
co-occurrence, correlation, cosine similarity); ¶ 38 (the type and number of
features are limitless and any type of feature can be used). Appellant’s
attorney argument that Agarwal does not determine weightings of webpage
features or scores of webpage features for advertisements is not persuasive
in view of these teachings. See Appeal Br. 25–26; Final Act. 14–15.
In view of Agarwal’s teaching of elements that correspond to the
claimed central feature advertisement, vertex webpage features, and edge
metrics, the Examiner reasonably determines a skilled artisan would have
been motivated to organize this data as a star schema as taught by Indukuri
to display the influence/relationship of different items/features on a central
feature/ad as claimed. Final Act. 6, 15–16; Ans. 8–9. Indukuri teaches this
advantage of a star topology. See Indukuri ¶¶ 2, 46, 53, 55.
Appeal 2020-001643
Application 13/746,582

35
In particular, Indukuri teaches star network topologies as an efficient
way to identify significant nodes in a plurality of nodes of a network based
on topology scores to influence another node(s) for business purposes such
as advertising placement. Indukuri ¶¶ 2, 22. The Examiner proposes to use
this teaching to achieve similar results with Agarwal’s network in which the
webpage features are scored/ranked to identify significant webpage features
that are likely to influence another node, namely, the central feature/ad node
in order to increase the probability of a click-through of the advertisement.
See KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398, 417 (2007) (holding the use
of a known technique to improve similar devices similarly is likely obvious
if the combination requires no more than ordinary skill in the art). Here, the
application of Indukuri’s teachings to Agarwal’s network requires ordinary
skill in the art as an artisan merely has to arrange the webpage features as a
plurality of vertices around a central node/ad with the scores determining the
relationship of each vertex node to the central ad node serving as “edges” as
claimed. Agarwal already determines webpage features and scores/ranks the
relationship of each feature to a central ad. Indukuri uses a star topology as
an efficient way to display/arrange this data in nodes and connecting edges.
The Examiner’s statement of a reason supported by a rational underpinning
to combine teachings of Agarwal and Indukuri to render obvious the claimed
method resolves the Appellant’s contention that the Examiner relied on
impermissible hindsight. See In re Cree, Inc., 818 F.3d 694, 702 n.3 (Fed.
Cir. 2016) (holding appellant’s hindsight argument was addressed by
showing that a proper motivation to combine the references in fact existed).
Accordingly, we sustain the rejection of claim 20 and claims 22, 25,
27, 28, and 30–41, which fall with claim 20.
Appeal 2020-001643
Application 13/746,582

36
Claim 21
Claim 21 depends from claim 20 and recites
wherein the metric for each edge between one of the vertices
and the central feature is determined by the analytics engine
based on a conversion rate of each of the same or similar
advertisements represented by the central feature when
displayed on one or more of the same or similar webpages
having the feature type corresponding to the one of the vertices.
Appeal Br. 35 (Claims App.).
Appellant argues that Agarwal teaches a predetermined feature of a
webpage that comprises a relevance score or cosine similarity function that
estimates the relationship of subject matter of a web page to subject matter
of a web advertisement. Id. at 29. Appellant argues that this feature is not a
metric for an edge based on a conversion rate of advertisements displayed on
a webpage having a feature type of the vertex. Id.
The Examiner has the better position. Agarwal determines metrics for
edges between a vertex and a central ad feature based on a conversion rate as
claimed. Agarwal matches the content of a webpage to web advertisements
to enhance or increase the likelihood of a user clicking on one of the web
advertisements. Agarwal ¶ 24. The amount of times that a user clicks on a
web advertisement relative to the number of times users visit a particular
web page where the advertisement is displayed is called the “click-through-
rate.” Id. The “click-through-rate” corresponds to the claimed conversion
rate, which refers to a percentage of users who click on an advertisement.
See Spec. ¶ 24. Agarwal uses expert statistic models to score the interaction
of webpage features and web advertisements to predict the probability of a
user clicking on a web advertisement on a webpage. Agarwal ¶ 41, id. ¶ 36
(predict the clickability of advertisement on a webpage).
Appeal 2020-001643
Application 13/746,582

37
Agarwal’s expert models provide a metric of the relationship between
webpage features and web advertisements to predict the likelihood that the
webpage feature and its relationship to an advertisement will result in a user
clicking on the web advertisement on that webpage. Id. ¶ 41. The expert
models and predicted probability are based on actual test data. Id. Expert
statistical models are tested with historical data corresponding to webpages
and web advertisements so that the predicted probability is based on actual
conversion rates from historical data corresponding to similar webpages and
advertisements as claimed. Id. ¶ 42. The models are modified continually
with new historical conversion data to ensure that the predicted probability
closely matches actual webpages and web advertisements for which a metric
is determined. Id.
Accordingly, we sustain the rejection of claim 21.
Claim 24
Appellant argues that Agarwal appears to disclose an expert model
that associates a particular webpage paired with a particular advertisement,
rather than multiple central features (advertisements) that belong to multiple
different classes as required by claim 24. Appeal Br. 30. Appellant argues
that Agarwal does not teach the ranking of vertices based on a probability
distribution of multiple classes over feature types represented by vertices to
rank the vertices based on a probability distribution of multiple classes over
feature types of vertices as required by claim 24. Id. at 30–31.
We agree with the Examiner that Agarwal teaches and suggests the
use of multiple web advertisements belonging to multiple classes and using
expert statistical models to create metrics that rank a probability distribution
of the webpage features to the different advertisement classes as claimed.
Appeal 2020-001643
Application 13/746,582

38
In this regard, Agarwal categorizes features and subject matter in
webpages and web advertisements as a “taxonomy.” Agarwal ¶¶ 18, 33; see
Ans. 19. Agarwal partitions groups of webpages and advertisements into
different categories such as sports, automobiles, and personal finance as
examples. Agarwal ¶ 34. Keywords are derived as features of webpages
and used to categorize webpages. Id. ¶ 25. Webpages can be ranked by
extracting keywords and categorizing the webpages. Id. ¶¶ 26, 28. Web
advertisements are categorized based on keywords appearing in the text of
the advertisements. Id. ¶ 34. Then, Agarwal applies a set of expert models
to the webpages and web advertisements to explain the interactions between
features on the webpages and features on the web advertisements. Id. ¶¶ 40,
41, Fig. 3 (steps 305, 310). The models are generated based on log data of
historical user interactions on webpages and with web advertisements and/or
actual test data/features that are used to predict a probability of a user
clicking on an advertisement based on features of multiple web pages and
web advertisements in multiple classes. Id. ¶¶ 40, 41; see Final Act. 25.
Agarwal considers webpage and central ad features belonging to multiple
classes and determines a probability distribution of the web page features.
Accordingly, we sustain the rejection of claim 24.
Claim 26
Appellant argues that Agarwal does not teach iteratively determining
a new probability of advertisement conversion for each feature type of each
class by altering a weight of each probability of advertisement conversion as
required by claim 26. Appeal Br. 31–32. As discussed above, Agarwal
continually modifies expert models to ensure that the models closely predict
a probability of a user clicking on a web advertisement. See Agarwal ¶ 42.
Appeal 2020-001643
Application 13/746,582

39
The Specification indicates only that an iterative probabilistic graph
learning technique can be used to identify features associated with a given
advertisement by collecting a feature dataset and transforming it into an
information network (star schema). Spec. ¶¶ 22, 55. Agarwal’s process of
continually updating expert models with historical data corresponding to the
webpages and advertisements iteratively modifies the expert models to
ensure that they closely predict a probability of a user clicking on a web
advertisement. Agarwal ¶ 42; see 37 C.F.R. § 41.37(c)(1)(iv) (arguments
that merely point out what a claim recites will not be considered argument
for the separate patentability of the claim); In re Lovin, 652 F.3d 1349, 1357
(Fed. Cir. 2011) (same).
Accordingly, we sustain the rejection of claim 26.
Claim 23
Rejected over Agarwal, Indukuri, and Gerace
Appellant does not present argument for the rejection of claim 23.
See Appeal Br. 21–32. Accordingly, we summarily sustain this rejection.
CONCLUSION
In summary:
Claims
Rejected
35 U.S.C.
§
Reference(s)/
Basis
Affirmed Reversed
20–28 30–41 101 Eligibility 20–28, 30–41
20–22, 24–
28, 30–41
103(a) Agarwal,
Indukuri
20–22, 24–
28, 30–41

23 103(a) Agarwal,
Indukuri,
Gerace
23
Overall
Outcome
20–28, 30–41
Appeal 2020-001643
Application 13/746,582

40
No time period for taking any subsequent action in connection with
this appeal may be extended under 37 C.F.R. § 1.136(a). See 37 C.F.R.
§ 1.136(a)(1)(iv).
AFFIRMED