DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This action is responsive to the original application filed on 6/18/2024. Acknowledgment is made with respect to a claim of priority to Provisional Application 62/945,791 filed on 12/9/2019. This is a Continuation Application of US Application number 17/1169,565 filed on 12/9/2020.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 2-5, 7, 14-15, and 19 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Regarding claim 2:
Claim 2 recites the limitation “wherein applying analysis of the user matchmaking embedding” in line 1. There is insufficient antecedent basis for this limitation in the claim because it is unclear what “the user matchmaking embedding” is being referred to. For the purposes of examination, this limitation is being interpreted as “wherein applying analysis of the user shared-item embedding”.
Dependent claims 3-5 and 7 are also rejected for inheriting the deficiencies of the base claim.
Regarding claim 14:
Claim 14 recites the limitation "The method of claim 16" in line 1. There is insufficient antecedent basis for this limitation in the claim because it is unclear what “method” is being referred to. For purposes of examination, this limitation is being interpreted as “The method of claim 1”.
Regarding claim 15:
Claim 15 recites the limitation "The method of claim 16" in line 1. There is insufficient antecedent basis for this limitation in the claim because it is unclear what “method” is being referred to. For purposes of examination, this limitation is being interpreted as “The method of claim 1”.
Regarding claim 17:
Claim 17 recites the limitation “wherein applying analysis of the user matchmaking embedding” in line 11. There is insufficient antecedent basis for this limitation in the claim because it is unclear what “the user matchmaking embedding” is being referred to. For the purposes of examination, this limitation is being interpreted as “wherein applying analysis of the user shared-item embedding”.
Regarding claim 19:
Claim 19 recites the limitation "The system of claim 1 " in line 1. There is insufficient antecedent basis for this limitation in the claim because it is unclear what “system” is being referred to. For the purposes of examination, this limitation is being interpreted as “The system of claim 18,”
Regarding claim 19:
Claim 19 also recites the limitation “wherein applying analysis of the user matchmaking embedding” in line 11. There is insufficient antecedent basis for this limitation in the claim because it is unclear what “the user matchmaking embedding” is being referred to. For the purposes of examination, this limitation is being interpreted as “wherein applying analysis of the user shared-item embedding”.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
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.
Claims 1-19 are rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The analysis of the claims will follow the 2019 Revised Patent Subject Matter Eligibility Guidance, 84 Fed. Reg. 50 (“2019 PEG”).
When considering subject matter eligibility under 35 U.S.C. 101, it must be determined whether the claim is directed to one of the four statutory categories of invention, i.e., process, machine, manufacture, or composition of matter (Step 1). If the claim does fall within one of the statutory categories, the second step in the analysis is to determine whether the claim is directed to a judicial exception (Step 2A). The Step 2A analysis is broken into two prongs. In the first prong (Step 2A, Prong 1), it is determined whether or not the claims recite a judicial exception (e.g., mathematical concepts, mental processes, certain methods of organizing human activity). If it is determined in Step 2A, Prong 1 that the claims recite a judicial exception, the analysis proceeds to the second prong (Step 2A, Prong 2), where it is determined whether or not the claims integrate the judicial exception into a practical application. If it is determined at step 2A, Prong 2 that the claims do not integrate the judicial exception into a practical application, the analysis proceeds to determining whether the claim is a patent-eligible application of the exception (Step 2B). If an abstract idea is present in the claim, any element or combination of elements in the claim must be sufficient to ensure that the claim integrates the judicial exception into a practical application, or else amounts to significantly more than the abstract idea itself.
Regarding claim 1:
Step 1 – Is the claim directed to a process, machine, manufacture, or composition of matter? – Yes, the claim is directed to a process.
Step 2A – Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon? – Yes, the claim recites the abstract ideas:
processing user data comprised of user feature data as input to a user neural network model and yielding a user embedding – This limitation is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]). The user neural network model is recited at a high level of generality which is merely using a computer as a tool to perform the concept.
processing the user embedding through a matchmaking neural network, which is a trained model to map user embeddings and content embeddings to a shared dimensional space, and yielding a user shared-item embedding – This limitation is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]). The matchmaking neural network is recited at a high level of generality which is merely using a computer as a tool to perform the concept.
applying analysis of the user shared-item embedding in selecting at least one content item associated with a content shared-item embedding within the matchmaking neural network – This limitation is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]).
Step 2A – Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application? The claim does not recite any additional limitations which integrate the abstract idea into a practical application. Specifically, the additional elements consist of “user neural network model” and “a matchmaking neural network”. The additional elements of “user neural network model” and “a matchmaking neural network” amount to generic computer components or models used as a tool to perform an existing process. Thus, the additional elements amount to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)). Thus, even when viewed individually and as an ordered combination, these additional elements do not integrate the abstract idea into a practical application, and the claim is thus directed to the abstract idea.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception? – The additional elements of “user neural network model” and “a matchmaking neural network” amount to generic computer components or models used as a tool to perform an existing process. Thus, the additional elements amount to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)). Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Regarding claim 2:
Step 2A – Prong 1 – The claim recites the additional abstract ideas:
wherein applying analysis of the user matchmaking embedding in selecting at least one content item comprises: calculating personalization scores between the user shared-item and content shared-item embeddings of a set of candidate content items – This limitation is further limiting the applying analysis limitation of claim 1, and is directed to a mathematical concept (mathematical formulas or equations [see MPEP 2106.04(a) B.]).
and updating prioritization of the set of candidate content items based in part on a calculated personalization scores – This limitation is further limiting the applying analysis limitation of claim 1, and is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]).
Thus, the judicial exception is not integrated into a practical application [see MPEP 2106.05(d) I.], failing step 2A prong 2. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B.
Regarding claim 3:
Step 2A – Prong 1 – The claim recites the additional abstract idea:
wherein calculating personalization scores between the user shared-item embedding and content shared-item embedding of a set of candidate content items comprises, for each content shared-item embedding of the set of candidate content items, calculating a personalization score by calculating a displacement between the user shared-item embedding and a content shared-item embeddings – This limitation is further limiting the calculating personalization scores limitation of claim 2, and is directed to a mathematical concept (mathematical formulas or equations [see MPEP 2106.04(a) B.]).
Thus, the judicial exception is not integrated into a practical application [see MPEP 2106.05(d) I.], failing step 2A prong 2. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B.
Regarding claim 4:
Step 2A – Prong 1 – The claim recites the additional abstract idea
wherein the displacement is the Euclidean distance between the user shared-item embedding to content shared-item embeddings – This limitation is further limiting the by calculating a displacement limitation from claim 3, and is directed to a mathematical concept (mathematical formulas or equations [see MPEP 2106.04(a) B.]).
Step 2A – Prong 2 – The claim recites the additional elements:
wherein the matchmaking neural network is a collaborative metric learning model – This limitation is directed to adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer [see MPEP 2106.05(f)].
Step 2B – The claim recites the additional elements:
wherein the matchmaking neural network is a collaborative metric learning model – This limitation is directed to adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer [see MPEP 2106.05(f)].
Thus, the judicial exception is not integrated into a practical application [see MPEP 2106.05(d) I.], failing step 2A prong 2. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B.
Regarding claim 5:
Step 2A – Prong 1 – The claim recites the additional abstract ideas:
wherein calculating personalization scores between the user shared-item embedding and a content shared-item embedding of the set of candidate content items comprises: generating a first classifier input by calculating a displacement between the user shared-item embedding and the content shared-item embedding – This limitation is further limiting the calculating personalization scores limitation from claim 2, and is directed to a mathematical concept (mathematical formulas or equations [see MPEP 2106.04(a) B.]).
generating at least a second classifier input by calculating a set of user-related content displacements between the content shared-item embedding and a set of user-related content shared-item embeddings – This limitation is further limiting the calculating personalization scores limitation from claim 2, and is directed to a mathematical concept (mathematical formulas or equations [see MPEP 2106.04(a) B.]).
processing the first classifier input and at least the second classifier input within a classifier model and outputting the personalization score – This limitation is further limiting the calculating personalization scores limitation from claim 2, and is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]).
Thus, the judicial exception is not integrated into a practical application [see MPEP 2106.05(d) I.], failing step 2A prong 2. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B.
Regarding claim 6:
Step 2A – Prong 1 -– The claim recites the additional abstract ideas:
calculating personalization scores between the user shared-item embedding and each content shared-item embedding of a set of candidate content items – This limitation is further limiting the applying analysis limitation of claim 1 and is directed to a mathematical concept (mathematical formulas or equations [see MPEP 2106.04(a) B.]).
and updating prioritization of the set of candidate content items based in part on the personalization scores – This limitation is further limiting the applying analysis limitation of claim 1 and is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]).
Step 2A – Prong 2 – The claim recites the additional elements:
wherein applying analysis of the user shared-item embedding comprises: receiving a query input – This limitation is further limiting the applying analysis limitation of claim 1, and is directed to insignificant extra-solution activity (merely data gathering [see MPEP 2106.05(g)]).
querying a content database to identify a filtered set of candidate content items – This limitation is directed to insignificant extra-solution activity [see MPEP 2106.05(g)].
Step 2B – The claim recites the additional elements:
wherein applying analysis of the user shared-item embedding comprises: receiving a query input – This limitation is further limiting the applying analysis limitation of claim 1, and is directed to receiving or transmitting data over a network, which the courts have recognized as well-understood, routine, conventional activity when they are claimed at a high level of generality or as insignificant extra-solution activity [see MPEP 2106.05(d) II. i.].
querying a content database to identify a filtered set of candidate content items – This limitation is further limiting the applying analysis limitation of claim 1, and is directed to storing and retrieving information in memory, which the courts have recognized as well-understood, routine, conventional activity when they are claimed at a high level of generality or as insignificant extra-solution activity [see MPEP 2106.05(d) II. iv.].
Thus, the judicial exception is not integrated into a practical application [see MPEP 2106.05(d) I.], failing step 2A prong 2. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B.
Regarding claim 7:
Step 2A – Prong 1 – The claim recites the additional abstract ideas:
wherein applying analysis of the user shared-item embedding further comprises grouping the set of candidate content items into a set of relevancy-groups – This limitation is further limiting the applying analysis limitation of claim 1, and is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]).
and wherein updating prioritization of the set of candidate content items based in part on the personalization scores comprises reprioritizing the set of candidate items by using personalization score to order candidate content items within the same relevancy group – This limitation is further limiting the updating prioritization limitation from claim 5, and is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]).
Thus, the judicial exception is not integrated into a practical application [see MPEP 2106.05(d) I.], failing step 2A prong 2. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B.
Regarding claim 8:
Step 2A – Prong 1 – The claim recites the additional abstract ideas:
identifying a set of candidate content items – This limitation is further limiting the applying analysis limitation of claim 1, and is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]).
calculating a personalization score between the user shared-item embedding and each content shared-item embeddings of the set of candidate content items – This limitation is further limiting the applying analysis limitation of claim 1, and is directed to a mathematical concept (mathematical formulas or equations [see MPEP 2106.04(a) B.]).
and updating prioritization of the set of candidate content items based in part on the calculated personalization scores – This limitation is further limiting the applying analysis limitation of claim 1, and is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]).
Step 2A – Prong 2 – The claim recites the additional element:
wherein applying analysis of the user shared-item embedding comprises: receiving a query input – This limitation is further limiting the applying analysis limitation of claim 1, and is directed to insignificant extra-solution activity (merely data gathering [see MPEP 2106.05(g)]).
Step 2B – The claim recites the additional element:
wherein applying analysis of the user shared-item embedding comprises: receiving a query input – This limitation is further limiting the applying analysis limitation of claim 1, and is directed to receiving or transmitting data over a network, which the courts have recognized as well-understood, routine, conventional activity when they are claimed at a high level of generality or as insignificant extra-solution activity [see MPEP 2106.05(d) II. i.].
Thus, the judicial exception is not integrated into a practical application [see MPEP 2106.05(d) I.], failing step 2A prong 2. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B.
Regarding claim 9:
Step 2A – Prong 1 – The claim recites the additional abstract ideas:
wherein identifying the set of candidate content items comprises identifying a set of shared-item embeddings satisfying a proximity condition relative to an anchor item – This limitation is further limiting the identifying a set of candidate content items limitation from claim 8, and is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]).
Thus, the judicial exception is not integrated into a practical application [see MPEP 2106.05(d) I.], failing step 2A prong 2. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B.
Regarding claim 10:
Step 2A – Prong 2 – The claim recites the additional element:
wherein the content shared-item embeddings are associated with product data records – This limitation is further limiting the content shared-item embeddings limitation from claim 1, and is directed to generally linking the use of an exception to a particular field of use [see MPEP 2106.5(h)].
Step 2B – The claim recites the additional element:
wherein the content shared-item embeddings are associated with product data records – This limitation is further limiting the content shared-item embeddings limitation from claim 1, and is directed to generally linking the use of an exception to a particular field of use [see MPEP 2106.5(h)].
Thus, the judicial exception is not integrated into a practical application [see MPEP 2106.05(d) I.], failing step 2A prong 2. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B.
Regarding claim 11:
Step 2A – Prong 2 – The claim recites the additional element:
wherein the content shared-item embeddings are associated with digital media content selected from the list of articles, images, video, and audio – This limitation is further limiting the content shared-item embeddings limitation from claim 1, and is directed to generally linking the use of an exception to a particular field of use [see MPEP 2106.5(h)].
Step 2B – The claim recites the additional element:
wherein the content shared-item embeddings are associated with digital media content selected from the list of articles, images, video, and audio – This limitation is further limiting the content shared-item embeddings limitation from claim 1, and is directed to generally linking the use of an exception to a particular field of use [see MPEP 2106.5(h)].
Thus, the judicial exception is not integrated into a practical application [see MPEP 2106.05(d) I.], failing step 2A prong 2. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B.
Regarding claim 12:
Step 2A – Prong 1 – The claim recites the additional abstract idea:
wherein selecting at least one content item associated with a content shared-item embedding within the matchmaking neural network comprises selecting a promotional content item – This limitation is further limiting the selecting at least one content item limitation from claim 1, and is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]).
Step 2A – Prong 2 – The claim recites the additional abstract idea:
and serving the promotional content item within a digital advertising network to a user – This limitation is further limiting the selecting at least one content item limitation from claim 1, and is directed to insignificant extra-solution activity [see MPEP 2106.05(g)].
Step 2B – The claim recites the additional element:
and serving the promotional content item within a digital advertising network to a user – This limitation is further limiting the selecting at least one content item limitation from claim 1, and is directed to receiving or transmitting data over a network, which the courts have recognized as well-understood, routine, conventional activity when they are claimed at a high level of generality or as insignificant extra-solution activity [see MPEP 2106.05(d) II. i.].
Thus, the judicial exception is not integrated into a practical application [see MPEP 2106.05(d) I.], failing step 2A prong 2. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B.
Regarding claim 13:
Step 2A – Prong 2 – The claim recites the additional element:
further comprising: training the user neural network – This limitation is further limiting the user neural network limitation from claim 1, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
and training the matchmaking neural network by applying collaborative metric learning – This limitation is further limiting the matchmaking neural network limitation from claim 1, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
Step 2B – The claim recites the additional element:
further comprising: training the user neural network – This limitation is further limiting the user neural network limitation from claim 1, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
and training the matchmaking neural network by applying collaborative metric learning – This limitation is further limiting the matchmaking neural network limitation from claim 1, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
Thus, the judicial exception is not integrated into a practical application [see MPEP 2106.05(d) I.], failing step 2A prong 2. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B.
Regarding claim 14:
Step 2A – Prong 1 – The claim recites the additional abstract idea:
for a set of content items: processing content data comprised of content feature data as input to the content neural network model and yielding a content embedding – This limitation is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]). The user neural network is recited at a high level of generality which is merely using a computer as a tool to perform the concept.
and processing the content embedding through the matchmaking neural network, yielding a content shared-item embedding – This limitation is further limiting the matchmaking neural network limitation from claim 1, and is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]). The matchmaking network is recited at a high level of generality which is merely using a computer as a tool to perform the concept.
Step 2A – Prong 2 – The claim recites the additional element:
further comprising training a content neural network – This limitation is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
Step 2B – The claim recites the additional element:
further comprising training a content neural network – This limitation is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
Thus, the judicial exception is not integrated into a practical application [see MPEP 2106.05(d) I.], failing step 2A prong 2. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B.
Regarding claim 15:
Step 2A – Prong 2 – The claim recites the additional elements:
wherein training the user neural network – This limitation is further limiting the user neural network limitation from claim 1, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
and training the matchmaking neural network comprises – This limitation is further limiting the matchmaking neural network limitation from claim 1, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
establishing a first set of user feature training data comprised of all user-associated interaction data and establishing a second set of user feature training data comprised of select set of user-associated interaction data – This limitation is further limiting the user neural network and the matchmaking neural network limitations from claim 1, and is directed to insignificant extra-solution activity (selecting a particular data source or type of data to be manipulated [see MPEP 2106.05(g)]).
and training the user neural network on the first set of user feature training data and the second set of user feature training data – This limitation is further limiting the matchmaking neural network limitations from claim 1, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
and training the matchmaking neural network by applying collaborative metric learning on using data derived from the first set of user feature training data and the second set of user feature training data – This limitation is further limiting the matchmaking neural network limitations from claim 1, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
Step 2B – The claim recites the additional elements:
wherein training the user neural network and training the matchmaking neural network comprises – This limitation is further limiting the user neural network and the matchmaking neural network limitations from claim 16, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
establishing a first set of user feature training data comprised of all user-associated interaction data and establishing a second set of user feature training data comprised of select set of user-associated interaction data – This limitation is further limiting the user neural network and the matchmaking neural network limitations from claim 1, and is directed to storing and retrieving information in memory, which the courts have recognized as well-understood, routine, conventional activity when they are claimed at a high level of generality or as insignificant extra-solution activity [see MPEP 2106.05(d) II. iv.].
and training the user neural network on the first set of user feature training data and the second set of user feature training data, and training the matchmaking neural network by applying collaborative metric learning on using data derived from the first set of user feature training data and the second set of user feature training data – This limitation is further limiting the user neural network and the matchmaking neural network limitations from claim 1, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
Thus, the judicial exception is not integrated into a practical application [see MPEP 2106.05(d) I.], failing step 2A prong 2. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B.
Regarding claim 16:
Step 1 – Is the claim directed to a process, machine, manufacture, or composition of matter? – Yes, the claim is directed to a non-transitory computer-readable medium.
Step 2A – Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon? – Yes, the claim recites the abstract idea:
processing user data comprised of user feature data as input to a user neural network model and yielding a user embedding – This limitation is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]). The user neural network model is recited at a high level of generality which is merely using a computer as a tool to perform the concept.
processing the user embedding through a matchmaking neural network, which is a trained model to map user embeddings and content embeddings to a shared dimensional space, and yielding a user shared-item embedding – This limitation is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]). The matchmaking neural network is recited at a high level of generality which is merely using a computer as a tool to perform the concept.
and applying analysis of the user shared-item embedding in selecting at least one content item associated with a content shared-item embedding within the matchmaking neural network – This limitation is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]).
Step 2A – Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application? – No, there are no additional elements that integrate the judicial exception into a practical application. The claim recites the additional element:
A non-transitory computer-readable medium storing instructions that, when executed by one or more processors of a computing platform, cause a computing platform to perform the operations – This limitation invokes a computer merely as a tool for performing an existing process [see MPEP 2106.05(f)(2)] and therefore fails to integrate the judicial exception into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception? -– No, there are no additional elements that amount to significantly more than the judicial exception. The claim recites the additional element:
A non-transitory computer-readable medium storing instructions that, when executed by one or more processors of a computing platform, cause a computing platform to perform the operations – This limitation invokes a computer merely as a tool for performing an existing process [see MPEP 2106.05(f)(2)] and therefore fails amount to significantly more than the judicial exception.
Regarding claim 17:
Step 2A – Prong 1 – The claim recites the additional abstract idea:
for a set of content items, processing content data comprised of content feature data as input to the content neural network and yielding a content embedding – This limitation is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]). The content neural network is recited at a high level of generality which is merely using a computer as a tool to perform the concept.
and processing the content embedding through the matchmaking neural network, yielding a content shared-item embedding – This limitation is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]). The matchmaking neural network is recited at a high level of generality which is merely using a computer as a tool to perform the concept.
wherein applying analysis of the user matchmaking embedding in selecting at least one content item comprises: calculating personalization scores between the user shared-item embedding and content shared-item embeddings of a set of candidate content items, – This limitation is further limiting the applying analysis limitation of claim 16, and is directed to a mathematical concept (mathematical formulas or equations [see MPEP 2106.04(a) B.]).
and updating prioritization of the set of candidate content items based in part on a calculated personalization score – This limitation is further limiting the applying analysis limitation of claim 16, and is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]).
Step 2A – Prong 2 – The claim also recites the additional elements:
further comprises instructions that cause the computing platform to perform the operations: training the user neural network – This limitation is further limiting the user neural network limitations of claim 16, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
training a content neural network – This limitation is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
training the matchmaking neural network by applying collaborative metric learning – This limitation is further limiting the matchmaking neural network limitations of claim 16, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
Step 2B – The claim also recites the additional elements:
further comprises instructions that cause the computing platform to perform the operations: training the user neural network – This limitation is further limiting the user neural network limitations of claim 16, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
training a content neural network – This limitation is further limiting the content neural network limitations of claim 16, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
training the matchmaking neural network by applying collaborative metric learning – This limitation is further limiting the matchmaking neural network limitations of claim 16, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
Thus, the judicial exception is not integrated into a practical application [see MPEP 2106.05(d) I.], failing step 2A prong 2. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B.
Regarding claim 18:
Step 1 – Is the claim directed to a process, machine, manufacture, or composition of matter? – No, the claim is not directed to a process, machine, manufacture, or composition of matter. The claim is directed to signals per se.
Step 2A – Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon? – Yes, the claim recites the abstract ideas:
processing user data comprised of user feature data as input to a user neural network model and yielding a user embedding – This limitation is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]). The user neural network model is recited at a high level of generality which is merely using a computer as a tool to perform the concept.
processing the user embedding through a matchmaking neural network, which is a trained model to map user embeddings and content embeddings to a shared dimensional space, and yielding a user shared-item embedding – This limitation is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]). The matchmaking neural network is recited at a high level of generality which is merely using a computer as a tool to perform the concept.
and applying analysis of the user shared-item embedding in selecting at least one content item associated with a content shared-item embedding within the matchmaking neural network – This limitation is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]).
Step 2A – Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application? – No, there are no additional elements that integrate the judicial exception into a practical application. The claim recites the additional elements:
A system comprising of: one or more computer-readable mediums storing instructions that, when executed by one or more computer processors, cause a computing platform to perform operations comprising – This limitation invokes a computer merely as a tool for performing an existing process [see MPEP 2106.05(f)(2)] and therefore fails to integrate the judicial exception into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception? -– No, there are no additional elements that amount to significantly more than the judicial exception. The claim recites the additional elements:
A system comprising of: one or more computer-readable mediums storing instructions that, when executed by one or more computer processors, cause a computing platform to perform operations comprising – This limitation invokes a computer merely as a tool for performing an existing process [see MPEP 2106.05(f)(2)] and therefore fails amount to significantly more than the judicial exception.
Regarding claim 19:
Step 2A – Prong 1 – The claim recites the additional abstract ideas:
for a set of content items, processing content data comprised of content feature data as input to the content neural network model and yielding a content embedding – This limitation is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]). The content neural network is recited at a high level of generality which is merely using a computer as a tool to perform the process.
and processing the content embedding through the matchmaking neural network, yielding a content shared-item embedding – This limitation is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]). The matchmaking neural network is recited at a high level of generality which is merely using a computer as a tool to perform the process.
and wherein applying analysis of the user matchmaking embedding in selecting at least one content item comprises: calculating personalization scores between the user shared-item embedding and content shared-item embeddings of a set of candidate content items – This limitation is further limiting the applying analysis limitation of claim 18, and is directed to a mathematical concept (mathematical formulas or equations [see MPEP 2106.04(a) B.])
and updating prioritization of the set of candidate content items based in part on a calculated personalization scores – This limitation is further limiting the applying analysis limitation of claim 18, and is directed to the abstract idea of a mental process (concepts performed in the human mind, including observation and evaluation [see MPEP 2106.04(a)(2) III. C.]).
Step 2A – Prong 2 – The claim recites the additional element:
wherein the instructions further cause the computing platform to perform the operations: training the user neural network – This limitation is further limiting the user neural network limitation of claim 18, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
training a content neural network – This limitation is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
training the matchmaking neural network by applying collaborative metric learning – This limitation is further limiting the matchmaking neural network limitation of claim 18, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
Step 2B – The claim recites the additional element:
wherein the instructions further cause the computing platform to perform the operations: training the user neural network – This limitation is further limiting the user neural network limitation of claim 18, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
training a content neural network – This limitation is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
training the matchmaking neural network by applying collaborative metric learning – This limitation is further limiting the matchmaking neural network limitations of claim 18, and is directed to adding the words “apply it” (or an equivalent) with the judicial exception, which merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)].
Thus, the judicial exception is not integrated into a practical application [see MPEP 2106.05(d) I.], failing step 2A prong 2. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B.
Claims 18-19 are further rejected under 35 U.S.C. § 101 because the claimed invention is directed towards non-statutory subject matter. The claims do not fall within at least one of the four categories of patent eligible subjected matter because the claimed invention is directed towards signals per se.
Regarding claim 18:
Claim 18 recites the term “computer-readable mediums storing instructions”, which is directed to a signal per se. “The computer- readable medium can be stored on any suitable computer readable media such as RAMs, ROMs, flash memory, EEPROMs, optical devices (CD or DVD), hard drives, floppy drives, or any suitable device.” [00207] does not exclude transitory forms of computer-readable mediums.
Dependent claim 19 is also rejected for inheriting the deficiencies of the base claim.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-4, 10-11, 13, and 14 are rejected under 35 U.S.C. 103 as being obvious over Zhang et al. (Zhang et al., “AdaCML: Adaptive Collaborative Metric Learning for Recommendation”, Apr. 24, 2019, Lecture Notes in Computer Science (LNISA,volume 11447), pp. 301-316, hereinafter “Zhang”) in view of Campo et al. (Campo et al., “Collaborative Metric Learning Recommendation System”, Mar. 1, 2018, arXiv:1803.00202, pp. 1-6, hereinafter “Campo”).
Regarding claim 1, Zhang discloses a method comprising:
processing user data comprised of user feature data [as input to a user neural network] and yielding a user embedding (Zhang: “As shown in Fig. 2, the bottom input layer consists of two sparse vectors uiU and vjI, respectively… For uiU, item IDs visited by user ui are represented as 1, and the unvisited items are 0” [*Examiner note: i.e., visited and unvisited items are being interpreted as user feature data, which is being used to represent the user as a vector (i.e., yielding a user embedding)] [Page 305, 3.1. The Model Architecture – Input Layer]);
processing the user embedding through a matchmaking neural network, which is a trained model to map user embeddings and content embeddings to a shared dimensional space, and yielding a user shared-item embedding (Zhang: “We use a memory component and an attention mechanism to flexibly acquire adaptive user representations based on candidate items” [*Examiner note: i.e., the adaptive user representation is being interpreted as a user shared-item embedding] [Page 303, 1. Introduction] … “Hence, the adaptive user representation uij is derived by weighting M according to the candidate item vj as:
PNG
media_image1.png
83
163
media_image1.png
Greyscale
, where … Ri+ denotes historical items liked by user ui” [Page 306, 3.1. The Model Architecture – Adaptive User Representation] … “
PNG
media_image2.png
559
754
media_image2.png
Greyscale
” [*Examiner note: (1): the user embedding uiU and (2): the item embedding vIj are mapped to a shared dimensional space, yielding (3): a user shared-item embedding uij] [Page 305, Fig. 2.]);
and applying analysis of the user shared-item embedding in selecting at least one content item associated with a content shared-item embedding within the matchmaking neural network (Zhang: “Ri− denotes the remaining items. Then, given user set U and item set I, for each user ui ∈ U, the recommendation task is defined as recommending a list of items from Ri− that the user ui may be interested in.” [*Examiner note: i.e., the process of recommending a list of items is being interpreted as applying analysis of the user shared-item embedding] [Page 304, 2. Problem Definition] … “For example, for user #541, the target item #1486 is a positive example in the testset, while items #856, #85, and #237 are historical ones. We can see that AdaCML assigns higher weights on items #856 and #85, a lower weight on item #237, successfully evaluating the target item #1486 as the item desired by the user.” [*Examiner note: i.e., evaluating the target item as the item desired by the user is being interpreted as selecting at least one content item associated with a positive example in the test set (i.e., associated with a content shared-item embedding)] [Page 313, 4.4. The Interpretability of AdaCML, Table 3]).
Zhang fails to explicitly disclose but Campo discloses processing user data comprised of user feature data as input to a user neural network and yielding a user embedding (Campo: “The plots, the meta data, and the customer data are inputs to the model.” [*Examiner note: i.e., the customer data is being interpreted as user feature data] [Page 7, 2. System Overview] … “we calculate, for each customer for each customer, the vector representing the purchase history of that customer” [*Examiner note: the customer history vector is being interpreted as a yielded user embedding] [Page 8, 4. System Implementation] … “we use a Siamese framework with two identical feed-forward multilayer neural networks” [Page 7, 3. Deep Metric Learning in the Product Space] … “
PNG
media_image3.png
782
850
media_image3.png
Greyscale
” [*Examiner note: (1): the customer history neural network of the Siamese framework is being interpreted as a user neural network)] [Page 6, Figure 1. Architecture of the prediction system]);
Zhang and Campo both disclose inventions relating to recommending content to a user and are therefore analogous. Zhang discloses a recommendation system that yields user embeddings and item embeddings. The user embeddings and item embeddings are mapped to a shared dimensional space to yield a user shared-item embedding. Campo discloses a recommendation system that uses a user neural network to generate the user embeddings. It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to generate user embeddings using a user neural network as disclosed by Campo. One would have been motivated to make this change because having a user neural network allows for the user vector to be adapted based on changes in customer preferences (Campo: “we reconstruct customer vectors based on past product purchases and a time discount function [13]. We do this to account for the fact that customers ‘drift’ in the product space because of changes in preferences and we would miss this important mechanism if we employed latent customer vectors” [Page 6, 1. Introduction]).
Regarding claim 2, Zhang in view of Campo discloses all of the limitations of claim 1 as shown in the rejection above. Zhang in view of Campo also discloses:
wherein applying analysis of the user matchmaking embedding in selecting at least one content item comprises: calculating personalization scores between the user shared-item embedding and content shared-item embeddings of a set of candidate content items (Zhang: “for each user ui ∈ U, the recommendation task is defined as recommending a list of items from Ri− that the user ui may be interested in.” [*Examiner note: i.e., the item set Ri− is being interpreted as a set of candidate content items] [Page 304, 2. Problem Definition] … “for an item vj, we formulate the item embedding vector as vj = QT vjI” [*Examiner note: i.e., the item embedding vector vj is being interpreted as a content shared-item embedding] [Page 306, 3.1. Model Architecture – Adaptive User Representation] … “the adaptive user representation uij is derived by weighting M according to the candidate item vj” [*Examiner note: i.e., the adaptive user representation is being interpreted as a user shared-item embedding] [Page 306, 3.1. The Model Architecture – Adaptive User Representation] … “when making predictions, we feed the final adaptive user representation uij and the target item embedding vj into the Euclidean distance: d(i, j) = || uij – vj- ||” [*Examiner note: i.e., Euclidean distance is being interpreted as the calculated personalization score between a user shared-item embedding and a content shared-item embedding] [Page 307, 3.1. Model Architecture – Prediction Layer]);
and updating prioritization of the set of candidate content items based in part on a calculated personalization scores (Zhang: “our model, like CML, also uses rank-based weighting scheme called Weighted Approximate-Rank Pairwise (WARP) loss, which was proposed by Weston et al. [16], to penalize positive items at a lower rank. As a result, we penalize the positive item vj based on its rank by setting σij = log(rankd(i, j) + 1), where rankd(i, j) denotes the rank of item vj in the recommended list of user ui.” [*Examiner note: i.e., the rank-based weighting is being interpreted as updating prioritization of the set of candidate content items] [Page 307, 3.2. Model Inference] … when making predictions, we feed the final adaptive user representation uij and the target item embedding vj into the Euclidean distance: d(i, j) = || uij – vj- ||” [*Examiner note: i.e., the recommended list is based in part on the Euclidean distance (i.e., the calculated personalization scores)] [Page 307, 3.1. Model Architecture – Prediction Layer]).
Regarding claim 3, Zhang in view of Campo discloses all of the limitations of claim 2 as shown in the rejection above. Zhang in view of Campo also discloses:
wherein calculating personalization scores between the user shared-item embedding and content shared-item embeddings of a set of candidate content items comprises, for each content shared-item embedding of the set of candidate content items, calculating a personalization score by calculating a displacement between the user shared-item embedding and a content shared-item embeddings (Zhang: “for each user ui ∈ U, the recommendation task is defined as recommending a list of items from Ri− that the user ui may be interested in.” [*Examiner note: i.e., the item set Ri− is being interpreted as a set of candidate content items] [Page 304, 2. Problem Definition] … “for an item vj, we formulate the item embedding vector as vj = QT vjI” [*Examiner note: i.e., the item embedding vector vj is being interpreted as a content shared-item embedding] [Page 306, 3.1. Model Architecture – Adaptive User Representation] … “the adaptive user representation uij is derived by weighting M according to the candidate item vj” [*Examiner note: i.e., the adaptive user representation is being interpreted as a user shared-item embedding] [Page 306, 3.1. The Model Architecture – Adaptive User Representation] … “when making predictions, we feed the final adaptive user representation uij and the target item embedding vj into the Euclidean distance: d(i, j) = || uij – vj- ||” [*Examiner note: i.e., Euclidean distance is being interpreted as the calculated personalization score between a user shared-item embedding and a content shared-item embedding] [Page 307, 3.1. Model Architecture – Prediction Layer]).
Regarding claim 4, Zhang in view of Campo discloses all of the limitations of claim 3 as shown in the rejection above. Zhang in view of Campo also discloses:
wherein the matchmaking neural network is a collaborative metric learning model (Zhang: “To this end, we propose a novel metric learning approach named Adaptive Collaborative Metric Learning (AdaCML) for recommendation.” [Page 301, Abstract])
and wherein the displacement is the Euclidean distance between the user shared-item embedding to content shared-item embeddings (Zhang: “when making predictions, we feed the final adaptive user representation uij and the target item embedding vj into the Euclidean distance: d(i, j) = || uij – vj- ||” [*Examiner note: i.e., d(i, j) is the calculated Euclidean distance (i.e., a calculated displacement)] [Page 307, 3.1. Model Architecture – Prediction Layer]).
Regarding claim 10, Zhang in view of Campo discloses all of the limitations of claim 1 as shown in the rejection above. Zhang in view of Campo also discloses:
wherein the content shared-item embeddings are associated with product data records (Zhang: “We will study the performance of AdaCML on the Amazon dataset [7, 13]. Amazon is an e-commerce platform and is widely used for product recommendation evaluation. We used two subsets of the Amazon review data as two datasets, namely Instant Video and Automotive… the Automotive dataset provides users’ purchase records” [*Examiner note: i.e., the item vectors (i.e., content shared-item embeddings) are associated with purchase records (i.e., product data records)] [Page 308, 4.1. Experimental Settings]).
Regarding claim 11, Zhang in view of Campo discloses all of the limitations of claim 1 as shown in the rejection above. Zhang in view of Campo also discloses:
wherein the content shared-item embeddings are associated with digital media content selected from the list of articles, images, video, and audio (Zhang: “We will study the performance of AdaCML on the Amazon dataset [7, 13]. Amazon is an e-commerce platform and is widely used for product recommendation evaluation. We used two subsets of the Amazon review data as two datasets, namely Instant Video and Automotive. The Instant Video dataset contains user ratings and review texts for the instant videos” [*Examiner note: i.e., the item vectors (i.e., content shared-item embeddings) are associated with video] [Page 308, 4.1. Experimental Settings]).
Regarding claim 13, Zhang in view of Campo discloses all of the limitations of claim 1 as shown in the rejection above. Zhang in view of Campo also discloses:
further comprising: training the user neural network (Campo: “To prepare the training dataset for the Siamese network, we use the customer purchase data to build the Cartesian product of the purchase history of each customer… To train the Siamese network, we append the fixed-length embedding vectors for each of the two movies in each row. We use the resulting dataset to train the Siamese network” [Page 8, 4. System Implementation] … “
PNG
media_image4.png
520
565
media_image4.png
Greyscale
” [*Examiner note: (1): the customer purchase data is input to the customer history neural network (i.e., the user neural network), therefore (2) training on the customer history trains the user neural network] [Page 6, Figure 1]);
Zhang and Campo both disclose inventions relating to recommending content to a user and are therefore analogous. Zhang discloses a recommendation system that yields user embeddings and item embeddings. The user embeddings and item embeddings are mapped to a shared dimensional space to yield a user shared-item embedding. Campo discloses a recommendation system that uses a user neural network to generate the user embeddings. It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to generate user embeddings using a user neural network as disclosed by Campo. One would have been motivated to make this change because training a user neural network allows for the user vector to be adapted based on changes in customer preferences (Campo: “we reconstruct customer vectors based on past product purchases and a time discount function [13]. We do this to account for the fact that customers ‘drift’ in the product space because of changes in preferences and we would miss this important mechanism if we employed latent customer vectors” [Page 6, 1. Introduction]).
training the matchmaking neural network by applying collaborative metric learning (Zhang: “After getting the adaptive user representation, we use the metric learning method to train and optimize our model AdaCML.” [*Examiner note: i.e., applying collaborative metric learning] [Page 305, 3.1. The Model Architecture]).
Regarding claim 14, Zhang in view of Campo discloses all of the limitations of claim 16 as shown in the rejection above. Zhang in view of Campo also discloses:
further comprising training a content neural network (Campo: “The first stage computes the high dimensional embedding of the movie plot … This first stage is trained on data for hundreds of movies” [Page 7, 2. System Overview] … “
PNG
media_image5.png
520
565
media_image5.png
Greyscale
” [*Examiner note: (1): the first stage handles movie plots, therefore (2) training the first stage trains the movie plot neural network (i.e., the content neural network).] [Page 6, Figure 1]);
for a set of content items: processing content data comprised of content feature data [as input to the content neural network model] and yielding a content embedding (Zhang: “As shown in Fig. 2, the bottom input layer consists of two sparse vectors uiU and vjI, that describe user ui and item vj respectively… For vjI, where the index corresponding where the index corresponding to item vj is represented as 1 and the rest of the positions are 0.” [*Examiner note: i.e., creating a sparse vector to represent item vj- is being interpreted as processing content data to yield a sparse vector ] [Page 305, 3.1. The Model Architecture – Input Layer])
processing content data comprised of content feature data as input to the content neural network model (Campo: “The plots, the meta data, and the customer data are inputs to the model … The first stage computes the high dimensional embedding of the movie plot (we use movie synopses and long movie scripts) and movie meta data (cast, movie, genre) in a dense space… The final dense representation of the movie plot and associated meta data is computed as the average of all the high-dimensional vectors for all the words and meta data elements for each movie.” [*Examiner note: i.e., the movie plot and meta data are being interpreted as content data, and the final dense representation of the movie plot and associated meta data is being interpreted as a content embedding] [Page 7, 2. System Overview] … “we use a Siamese framework with two identical feed-forward multilayer neural networks” [Page 7, 3. Deep Metric Learning in the Product Space] … “
PNG
media_image6.png
685
747
media_image6.png
Greyscale
” [*Examiner note: (1): the movie plot neural network of the Siamese architecture is being interpreted as a content neural network)] [Page 6, Figure 1]).
Zhang and Campo both disclose inventions relating to recommending content to a user and are therefore analogous. Zhang discloses a recommendation system that yields user embeddings and item embeddings. The user embeddings and item embeddings are mapped to a shared dimensional space to yield a content shared-item embedding. Campo discloses a recommendation system that uses a content neural network to generate the item embeddings. It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to generate item embeddings using a content neural network as disclosed by Campo. One would have been motivated to make this change because having a content neural network allows the item vector to place more weight on more relevant aspects of the item (Campo: “In addition to the plots, this step also requires collecting meta data about those movies, like movie genre, casting, etc… We do this because we want to place as much weight as possible on the plot description and the cast. The plots, the meta data, and the customer data are inputs to the model” [Page 3, 2. System Overview]).
and processing the content embedding through the matchmaking neural network, yielding a content shared-item embedding (Zhang: “each dense representation can be regarded as the latent feature vector for an item. Hence, for an item vj, we formulate the item embedding vector as vj = QT vjI, where Q ∈ ℝK×M, denoting the latent vector matrix for all items. M is the number of items and K is the dimensionality of the user/item latent vector.” [*Examiner note: i.e., the item embedding vector vj is being interpreted as a content shared-item embedding] [Page 306, 3.1. Model Architecture – Adaptive User Representation]);
Claims 5 and 7 are rejected under 35 U.S.C. 103 as being obvious over Zhang in view of Campo, and in further view of Tan et al. (Tan et al., “A Collaborative Filtering Recommendation Algorithm Based on Item Classification”, Sep. 4, 2009, 2009 Pacific-Asia Conference on Circuits,Communications and System, pp. 1-4, hereinafter “Tan”).
Regarding claim 5, Zhang in view of Campo discloses all of the limitations of claim 2 as shown in the rejection above. Zhang in view of Campo fails to disclose:
wherein calculating personalization scores between the user shared-item embedding and a content shared-item embedding of the set of candidate content item comprises: generating a first classifier input by calculating a displacement between the user shared-item embedding and the content shared-item embedding;
generating at least a second classifier input by calculating a set of user-related content displacements between the content shared-item embedding and a set of user-related content shared-item embeddings;
processing the first classifier input and at least the second classifier input within a classifier model and outputting the personalization score.
However, Tan discloses:
wherein calculating personalization scores between the user shared-item embedding and a content shared-item embedding of the set of candidate content item comprises: generating a first classifier input by calculating a displacement between the user shared-item embedding and the content shared-item embedding (Tan: “we propose a collaborative filtering recommendation algorithm based on the item classification” [Page 1, Col. 2] … “The rating of the target user u to the target item t is as following:
PNG
media_image7.png
144
374
media_image7.png
Greyscale
Where Rui is the rating of the target user u to the nearest item i” [*Examiner note: (1): Rui is a first classifier input that is a displacement between user u (i.e., a user shared-item embedding)] and the nearest item i (i.e., the content shared item embedding)] [Page 2, B. Prediction using item-based CF]);
generating at least a second classifier input by calculating a set of user-related content displacements between the content shared-item embedding and a set of user-related content shared-item embeddings (Tan: “we propose a collaborative filtering recommendation algorithm based on the item classification” [Page 1, I. Introduction] … “The rating of the target user u to the target item t is as following:
PNG
media_image8.png
141
374
media_image8.png
Greyscale
… sim(t, i) is the similarity of the target item t and the neighbor item i” [*Examiner note: (1): sim(t, i) is a second classifier input, and is a displacement between a target item t (i.e., the content shared-item embedding) and neighbor items i (i.e., a set of user-related content shared-item embedding)] [Page 2, B. Prediction using item-based CF]);
processing the first classifier input and at least the second classifier input within a classifier model and outputting the personalization score (Tan: “this paper proposes a collaborative filtering recommendation algorithm based on the item classification to pre-produce the ratings” [*Examiner note: i.e., this is a classifier model] … “The rating of the target user u to the target item t is as following:
PNG
media_image9.png
140
374
media_image9.png
Greyscale
“ [*Examiner note: i.e., Put is the outputted personalization score] [Page 2, B. Prediction using item-based CF]).
Zhang in view of Campo and Tan both disclose inventions relating to providing content recommendations to a user and are therefore analogous. It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the calculating of personalization scores taught Zhang in view of Campo in order to process the calculated classifier inputs as taught by Tan. One having ordinary skill in the art would have been motivated to make this change before the effective filing date of the claimed invention because having multiple classifier inputs helps reduce the issue of data sparsity in the model (Tan: “The collaborative filtering recommendation method based on item classification prediction can alleviate the sparsity problem of the user-item rating dataset, and can provide better recommendation than traditional collaborative filtering.” [Page 1, I. Introduction]).
Regarding claim 7, Zhang in view of Campo and in further view of Tan discloses all of the limitations of claim 5 as shown in the rejection above. Zhang in view of Campo and in further view of Tan also discloses:
wherein applying analysis of the user shared-item embedding further comprises grouping the set of candidate content items into a set of relevancy-groups (Zhang: “For each positive user-item pair, (ui, vj ), we randomly select several items from the user’s unvisited records to form several negative user-item pairs. Similar to positive examples, negative pairs are represented as corresponding users and items hidden vectors in the same way. … S is a set of positive user-item pairs” [*Examiner note: i.e., the set of positive user-item pairs and the set of negative user-item pairs are being interpreted as relevancy groups] [Page 307, 3.2. Model Inference]);
and wherein updating prioritization of the set of candidate content items based in part on the personalization scores comprises reprioritizing the set of candidate content items by using personalization score to order candidate content items within the same relevancy group (Zhang: “our model, like CML, also uses rank-based weighting scheme called Weighted Approximate-Rank Pairwise (WARP) loss, which was proposed by Weston et al. [16], to penalize positive items at a lower rank. As a result, we penalize the positive item vj based on its rank by setting σij = log(rankd(i, j) + 1), where rankd(i, j) denotes the rank of item vj in the recommended list of user ui.” [*Examiner note: i.e., the rank-based weighting is being interpreted as updating prioritization of the content items within the set of positive user-item pairs (i.e., within the same relevancy group)] [Page 307, 3.2. Model Inference]).
Claims 6, 8, and 9 are rejected under 35 U.S.C. 103 as being obvious over Zhang in view of Campo, and in further view of Farrelly (US 20150074090 A1, hereinafter “Farrelly”).
Regarding claim 6, Zhang in view of Campo discloses all of the limitations of claim 1 as shown in the rejection above. Zhang in view of Campo also discloses:
calculating personalization scores between the user shared-item embedding and content shared-item embeddings of a set of candidate content items (Zhang: “for each user ui ∈ U, the recommendation task is defined as recommending a list of items from Ri− that the user ui may be interested in.” [*Examiner note: i.e., the item set Ri− is being interpreted as a set of candidate content items] [Page 304, 2. Problem Definition] … “for an item vj, we formulate the item embedding vector as vj = QT vjI” [*Examiner note: i.e., the item embedding vector vj is being interpreted as a content shared-item embedding] [Page 306, 3.1. Model Architecture – Adaptive User Representation] … “the adaptive user representation uij is derived by weighting M according to the candidate item vj” [*Examiner note: i.e., the adaptive user representation is being interpreted as a user shared-item embedding] [Page 306, 3.1. The Model Architecture – Adaptive User Representation] … “when making predictions, we feed the final adaptive user representation uij and the target item embedding vj into the Euclidean distance: d(i, j) = || uij – vj- ||” [*Examiner note: i.e., Euclidean distance is being interpreted as the calculated personalization score between a user shared-item embedding and a content shared-item embedding] [Page 307, 3.1. Model Architecture – Prediction Layer]);
and updating prioritization of the set of candidate content items based in part on a calculated personalization scores (Zhang: “our model, like CML, also uses rank-based weighting scheme called Weighted Approximate-Rank Pairwise (WARP) loss, which was proposed by Weston et al. [16], to penalize positive items at a lower rank. As a result, we penalize the positive item vj based on its rank by setting σij = log(rankd(i, j) + 1), where rankd(i, j) denotes the rank of item vj in the recommended list of user ui.” [*Examiner note: i.e., the rank-based weighting is being interpreted as updating prioritization of the set of candidate content items] [Page 307, 3.2. Model Inference] … when making predictions, we feed the final adaptive user representation uij and the target item embedding vj into the Euclidean distance: d(i, j) = || uij – vj- ||” [*Examiner note: i.e., the recommended list is based in part on the Euclidean distance (i.e., the calculated personalization scores)] [Page 307, 3.1. Model Architecture – Prediction Layer]).
Zhang in view of Campo fails to disclose:
wherein applying analysis of the user shared-item embedding comprises: receiving a query input, querying a content database to identify a filtered set of candidate content items,
However, Farrelly discloses:
wherein applying analysis of the user shared-item embedding comprises: receiving a query input, querying a content database to identify a filtered set of candidate content items (Farrelly: “the music engine 20 sends a request to the recommendation service 26 such that the music database 28 is queried to identify other songs historically related to the seed song” [*Examiner note: i.e., the music database is being interpreted as a content database that receives a query to identify related songs (i.e., a filtered set of candidate content items] [0020] [FIG. 1] … “At this point, the central server 12, and more specifically the recommendation engine 18, generates a list of recommended songs based on the responses from the servers 30, 36, and 42 (step 122).” [*Examiner note: i.e., as part of applying analysis of the user shared-item] [0032] [FIG. 2]),
Zhang in view of Campo and Farrelly both disclose inventions relating to providing content recommendations to a user and are therefore analogous. It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the analysis of the user shared-item embedding as taught by Zhang in view of Campo to include querying a database for a set of candidate content items as disclosed by Farrelly. One having ordinary skill in the art would have been motivated to make this change before the effective filing date of the claimed invention because this set of content items may provide insight onto the tastes and interests of the user (Farrelly: “Further, their musical tastes are also influenced by the type and nature of other media (such as movies and television) that were popular at particular points in time.” [0006]).
Regarding claim 8, Zhang in view of Campo discloses all of the limitations of claim 1 as shown in the rejection above. Zhang in view of Campo also discloses:
calculating personalization scores between the user shared-item embedding and content shared-item embeddings of a set of candidate content items (Zhang: “for each user ui ∈ U, the recommendation task is defined as recommending a list of items from Ri− that the user ui may be interested in.” [*Examiner note: i.e., the item set Ri− is being interpreted as a set of candidate content items] [Page 304, 2. Problem Definition] … “for an item vj, we formulate the item embedding vector as vj = QT vjI” [*Examiner note: i.e., the item embedding vector vj is being interpreted as a content shared-item embedding] [Page 306, 3.1. Model Architecture – Adaptive User Representation] … “the adaptive user representation uij is derived by weighting M according to the candidate item vj” [*Examiner note: i.e., the adaptive user representation is being interpreted as a user shared-item embedding] [Page 306, 3.1. The Model Architecture – Adaptive User Representation] … “when making predictions, we feed the final adaptive user representation uij and the target item embedding vj into the Euclidean distance: d(i, j) = || uij – vj- ||” [*Examiner note: i.e., Euclidean distance is being interpreted as the calculated personalization score between a user shared-item embedding and a content shared-item embedding] [Page 307, 3.1. Model Architecture – Prediction Layer]);
and updating prioritization of the set of candidate content items based in part on a calculated personalization scores (Zhang: “our model, like CML, also uses rank-based weighting scheme called Weighted Approximate-Rank Pairwise (WARP) loss, which was proposed by Weston et al. [16], to penalize positive items at a lower rank. As a result, we penalize the positive item vj based on its rank by setting σij = log(rankd(i, j) + 1), where rankd(i, j) denotes the rank of item vj in the recommended list of user ui.” [*Examiner note: i.e., the rank-based weighting is being interpreted as updating prioritization of the set of candidate content items] [Page 307, 3.2. Model Inference] … when making predictions, we feed the final adaptive user representation uij and the target item embedding vj into the Euclidean distance: d(i, j) = || uij – vj- ||” [*Examiner note: i.e., the recommended list is based in part on the Euclidean distance (i.e., the calculated personalization scores)] [Page 307, 3.1. Model Architecture – Prediction Layer]).
Zhang in view of Campo fails to disclose:
wherein applying analysis of the user shared-item embedding comprises: receiving a query input, identifying a set of candidate content items
However, Farrelly discloses:
wherein applying analysis of the user shared-item embedding comprises: receiving a query input, identifying a set of candidate content items (Farrelly: “the music engine 20 sends a request to the recommendation service 26 such that the music database 28 is queried to identify other songs historically related to the seed song” [*Examiner note: i.e., the music database is being interpreted as a content database that receives a query to identify related songs (i.e., a set of candidate content items] [0020] [FIG. 1] … “At this point, the central server 12, and more specifically the recommendation engine 18, generates a list of recommended songs based on the responses from the servers 30, 36, and 42 (step 122).” [*Examiner note: i.e., as part of applying analysis of the user shared-item] [0032] [FIG. 2]),
Zhang in view of Campo and Farrelly both disclose inventions relating to providing content recommendations to a user and are therefore analogous. It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the analysis of the user shared-item embedding as taught by Zhang in view of Campo to include querying a database for a set of candidate content items as disclosed by Farrelly. One having ordinary skill in the art would have been motivated to make this change before the effective filing date of the claimed invention because this set of content items may provide insight onto the tastes and interests of the user (Farrelly: “Further, their musical tastes are also influenced by the type and nature of other media (such as movies and television) that were popular at particular points in time.” [0006]).
Regarding claim 9, Zhang in view of Campo, and in further view of Farrelly discloses all of the limitations of claim 8 as shown in the rejection above. Zhang in view of Campo, and in further view of Farrelly also discloses:
wherein identifying the set of candidate content items comprises identifying a set of shared-item embeddings satisfying a proximity condition relative to an anchor item (Farrelly: “the recommendation service 26 queries the music database 28 based on the time window for the seed song and sends a response including a list of songs historically related to the seed song to the central server 12 (step 112). The songs returned by the recommendation service 26 of the music server 30 may include, for example, songs released during the time window for the seed song, songs that were on one or more music charts during the time window for the seed song, songs by artists that had a song on one or more music charts during the time window for the seed song, songs by artists who released a song or album during the time window for the seed song, or the like” [*Examiner note: i.e., songs that satisfy the conditions for being related to the seed song are satisfying a proximity condition relative to the seed song, and the seed song is being interpreted as an anchor item] [0027]).
Zhang in view of Campo and Farrelly both disclose inventions relating to providing content recommendations to a user and are therefore analogous. It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the analysis of the user shared-item embedding taught by Zhang in view of Campo to include querying a database for a set of candidate content items as disclosed by Farrelly. One having ordinary skill in the art would have been motivated to make this change before the effective filing date of the claimed invention because this set of content items may provide insight onto the tastes and interests of the user (Farrelly: “Further, their musical tastes are also influenced by the type and nature of other media (such as movies and television) that were popular at particular points in time.” [0006]).
Claim 12 is rejected under 35 U.S.C. 103 as being obvious over Zhang in view of Campo in view of Jacobi et al. (US 20010021914 A1, hereafter “Jacobi”).
Regarding claim 12, Zhang in view of Campo discloses all of the limitations of claim 1, as shown in the rejection above. Zhang in view of Campo fails to disclose:
wherein selecting at least one content item associated with a content shared-item embedding within the matchmaking neural network comprises selecting a promotional content item and serving the promotional content item within a digital advertising network to a user
However, Jacobi discloses:
wherein selecting at least one content item associated with a content shared-item embedding within the matchmaking neural network comprises selecting a promotional content item and serving the promotional content item within a digital advertising network to a user (Jacobi: “Finally, in step 94, a list of the top M (e.g., 15) items of the recommendations list are returned to the Web server 32 (FIG. 1). The Web server incorporates this list into one or more Web pages that are returned to the user, with each recommended item being presented as a hypertextual link to the item's product information page. The recommendations may alternatively be conveyed to the user by email, facsimile, or other transmission method. Further, the recommendations could be presented as advertisements for the recommended items.” [0072]).
Zhang in view of Campo and Jacobi both disclose inventions relating to providing content recommendations to user and are therefore analogous. It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to implement the providing of advertisements for a list of selected items as taught by Jacobi onto the selected items taught by Zhang in view of Campo. One having ordinary skill in the art would have been motivated to make this change before the effective filing date of the claimed invention because the promotional content can be generated without direct input from the user regarding their interests (Jacobi: “An important benefit of the service is that the recommendations are generated without the need for the user, or any other users, to rate items.” [0011]).
Claim 15 is rejected under 35 U.S.C. 103 as being unpatentable over Zhang in view of Campo and in further view of Wei-Cheng et al. (US 11,245,665 B2, hereinafter “Wei-Cheng”).
Regarding claim 15, Zhang in view of Campo discloses all of the limitations of claim 16 as shown in the rejection above. Zhang in view of Campo fails to disclose:
wherein training the user neural network and training the matchmaking neural network comprises: establishing a first set of user feature training data comprised of all user-associated interaction data
and establishing a second set of user feature training data comprised of select set of user-associated interaction data,
and training the user neural network on the first set of user feature training data and the second set of user feature training data,
and training the matchmaking neural network by applying collaborative metric learning on using data derived from the first set of user feature training data and the second set of user feature training data.
However, Wei-Cheng discloses:
wherein training the user neural network and training the matchmaking neural network comprises: establishing a first set of user feature training data comprised of all user-associated interaction data (Wei-Cheng: “domain name searches, selected suggested domain names and domain names registered to the user are stored as the training data in a training database.” [*Examiner note: the training data is user-associated interaction data] [Abstract] … “The training process of the learning algorithm may be repeated using all (or some given portion) of the training data. A full-cycle of using all (or some desired portion) of the training data may be referred to as an epoch. Additional epochs, i.e., full-cycles of using all (or some given portion) of the training data, may be performed until the learning algorithm has been trained to a desired level.” [*Examiner note: i.e., a full cycle using all of the training data is a first set of user feature training data comprised of all user-associated interaction data] [Col. 3, 28 – 34])
and establishing a second set of user feature training data comprised of select set of user-associated interaction data (Wei-Cheng: “If parts of the training data are not used for training, these unused parts of the training data may be used to check the progress of the training of the learning algorithm to determine if the learning algorithm has been trained to a desired level.” [*Examiner note: i.e., a second set of select user-associated interaction data] [Col. 3, lines 34 – 38]),
and training the user neural network on the first set of user feature training data and the second set of user feature training data (Wei-Cheng: “The training process of the learning algorithm may be repeated using all (or some given portion) of the training data. A full-cycle of using all (or some desired portion) of the training data may be referred to as an epoch. Additional epochs, i.e., full-cycles of using all (or some given portion) of the training data, may be performed until the learning algorithm has been trained to a desired level. If parts of the training data are not used for training, these unused parts of the training data may be used to check the progress of the training of the learning algorithm to determine if the learning algorithm has been trained to a desired level.” [*Examiner note: i.e., training on data derived from the first set and second set of user training data, which can be applied to a neural network (i.e., the user neural network)] [Col. 3, lines 28 – 38]),
and training the matchmaking neural network by applying collaborative metric learning on using data derived from the first set of user feature training data and the second set of user feature training data (Wei-Cheng: “The training process of the learning algorithm may be repeated using all (or some given portion) of the training data. A full-cycle of using all (or some desired portion) of the training data may be referred to as an epoch. Additional epochs, i.e., full-cycles of using all (or some given portion) of the training data, may be performed until the learning algorithm has been trained to a desired level. If parts of the training data are not used for training, these unused parts of the training data may be used to check the progress of the training of the learning algorithm to determine if the learning algorithm has been trained to a desired level.” [*Examiner note: i.e., the training can be applied to a neural network (i.e., a matchmaking neural network)] [Col. 3, lines 28 – 38]).
Zhang in view of Campo and Wei-Cheng both disclose inventions relating to providing recommended content to a user and are therefore analogous. It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the training of the user and matchmaking networks taught by Zhang in view of Campo to incorporate the training on full and partial datasets taught by Wei-Cheng. One having ordinary skill in the art would have been motivated to make this change because training helps increase the accuracy of future predictions of the model (Wei-Cheng: “Using any desired technique, such as, as non-limiting examples, gradient descent and/or back propagation, the parameters of the encoder of the learning algorithm are adjusted so that the input is more likely to produce the expected output in future runs.” [Col. 3, lines 23 – 27]).
Claims 16-19 are rejected under 35 U.S.C. 103 as being obvious over Zhang in view of Campo, and further in view of Tareen et al. (US 8,548,996 B2, hereinafter “Tareen”).
Regarding claim 16, Zhang discloses:
processing user data comprised of user feature data [as input to a user neural network] and yielding a user embedding (Zhang: “As shown in Fig. 2, the bottom input layer consists of two sparse vectors uiU and vjI, respectively… For uiU, item IDs visited by user ui are represented as 1, and the unvisited items are 0” [*Examiner note: i.e., visited and unvisited items are being interpreted as user feature data, which is being used to represent the user as a vector (i.e., yielding a user embedding)] [Page 305, 3.1. The Model Architecture – Input Layer]);
processing the user embedding through a matchmaking neural network, which is a trained model to map user embeddings and content embeddings to a shared dimensional space, and yielding a user shared-item embedding (Zhang: “we propose a novel metric learning approach named Adaptive Collaborative Zhang: “We use a memory component and an attention mechanism to flexibly acquire adaptive user representations based on candidate items” [*Examiner note: i.e., the adaptive user representation is being interpreted as a user shared-item embedding] [Page 303, 1. Introduction] … “Hence, the adaptive user representation uij is derived by weighting M according to the candidate item vj as:
PNG
media_image1.png
83
163
media_image1.png
Greyscale
, where … Ri+ denotes historical items liked by user ui” [Page 306, 3.1. The Model Architecture – Adaptive User Representation] … “
PNG
media_image2.png
559
754
media_image2.png
Greyscale
” [*Examiner note: (1): the user embedding uiU and (2): the item embedding vIj are mapped to a shared dimensional space, yielding (3): a user shared-item embedding uij] [Page 305, Fig. 2.]);
and applying analysis of the user shared-item embedding in selecting at least one content item associated with a content shared-item embedding within the matchmaking neural network (Zhang: “Ri− denotes the remaining items. Then, given user set U and item set I, for each user ui ∈ U, the recommendation task is defined as recommending a list of items from Ri− that the user ui may be interested in.” [*Examiner note: i.e., the process of recommending a list of items is being interpreted as applying analysis of the user shared-item embedding] [Page 304, 2. Problem Definition] … “For example, for user #541, the target item #1486 is a positive example in the testset, while items #856, #85, and #237 are historical ones. We can see that AdaCML assigns higher weights on items #856 and #85, a lower weight on item #237, successfully evaluating the target item #1486 as the item desired by the user.” [*Examiner note: i.e., evaluating the target item as the item desired by the user is being interpreted as selecting at least one content item associated with a positive example in the test set (i.e., associated with a content shared-item embedding)] [Page 313, 4.4. The Interpretability of AdaCML, Table 3]).
Zhang fails to disclose:
A non-transitory computer-readable medium storing instructions that, when executed by one or more computer processors of a computing platform, cause a computing platform to perform the operations,
processing user data comprised of user feature data as input to a user neural network
However, Campo discloses:
processing user data comprised of user feature data as input to a user neural network (Campo: “The plots, the meta data, and the customer data are inputs to the model.” [*Examiner note: i.e., the customer data is being interpreted as user feature data] [Page 7, 2. System Overview] … “we use a Siamese framework with two identical feed-forward multilayer neural networks” [Page 7, 3. Deep Metric Learning in the Product Space] … “we calculate, for each customer for each customer, the vector representing the purchase history of that customer” [*Examiner note: the customer history vector is being interpreted as a yielded user embedding] [Page 8, 4. System Implementation] … “
PNG
media_image3.png
782
850
media_image3.png
Greyscale
” [*Examiner note: (1): the customer history neural network of the Siamese framework is being interpreted as a user neural network)] [Page 6, Figure 1. Architecture of the prediction system]);
Zhang and Campo both disclose inventions relating to recommending content to a user and are therefore analogous. Zhang discloses a recommendation system that yields user embeddings and item embeddings. The user embeddings and item embeddings are mapped to a shared dimensional space to yield a user shared-item embedding. Campo discloses a recommendation system that uses a user neural network to generate the user embeddings. It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to generate user embeddings using a user neural network as disclosed by Campo. One would have been motivated to make this change because having a user neural network allows for the user vector to be adapted based on changes in customer preferences (Campo: “we reconstruct customer vectors based on past product purchases and a time discount function [13]. We do this to account for the fact that customers ‘drift’ in the product space because of changes in preferences and we would miss this important mechanism if we employed latent customer vectors” [Page 2, 1. Introduction]).
Zhang in view of Campo fails to disclose:
A non-transitory computer-readable medium storing instructions that, when executed by one or more computer processors of a computing platform, cause a computing platform to perform the operations:
However, Tareen discloses:
A non-transitory computer-readable medium storing instructions that, when executed by one or more computer processors of a computing platform, cause a computing platform to perform the operations (Tareen: “a computer program product embodied on a computer readable storage medium; and/or a processor, such as a processor configured to execute instructions stored on and/or provided by a memory coupled to the processor.” [Col. 1, lines 55 – 58]):
Zhang in view of Campo and Tareen both disclose inventions relating to recommending content to a user and are therefore analogous. It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to render the teachings of Zhang in view of Campo executable by a computer by storing instructions onto computer-readable media using the teachings of Tareen. One having ordinary skill in the art would have been motivated to make this change because computers can be adapted and specialized for specific tasks (Tareen: “Unless stated otherwise, a component such as a processor or a memory described as being configured to perform a task may be implemented as a general component that is temporarily configured 65 to perform the task at a given time or a specific component that is manufactured to perform the task.” [Col. 1, lines 62 – 67]).
Regarding claim 17, Zhang in view of Campo and in further view of Tareen discloses all of the limitations of claim 16 as shown in the rejection above. Zhang in view of Campo and in further view of Tareen also discloses:
further comprises instructions that cause the computing platform to perform the operations: training the user neural network (Campo: “To prepare the training dataset for the Siamese network, we use the customer purchase data to build the Cartesian product of the purchase history of each customer… To train the Siamese network, we append the fixed-length embedding vectors for each of the two movies in each row. We use the resulting dataset to train the Siamese network” [Page 8, 4. System Implementation] … “
PNG
media_image4.png
520
565
media_image4.png
Greyscale
” [*Examiner note: (1): the customer purchase data is input to the customer history neural network (i.e., the user neural network), therefore (2) training on the customer history trains the user neural network] [Page 6, Figure 1]);
training a content neural network (Campo: “The first stage computes the high dimensional embedding of the movie plot … This first stage is trained on data for hundreds of movies” [Page 3, 2. System Overview] … “
PNG
media_image5.png
520
565
media_image5.png
Greyscale
” [*Examiner note: (1): the first stage handles movie plots, therefore (2) training the first stage trains the movie plot neural network (i.e., the content neural network).] [Page 2, Figure 1]);
and training the matchmaking neural network by applying collaborative metric learning (Zhang: “After getting the adaptive user representation, we use the metric learning method to train and optimize our model AdaCML.” [*Examiner note: i.e., applying collaborative metric learning] [Page 305, 3.1. The Model Architecture]).
for a set of content items: processing content data comprised of content feature data [as input to the content neural network model] and yielding a content embedding (Zhang: “As shown in Fig. 2, the bottom input layer consists of two sparse vectors uiU and vjI, that describe user ui and item vj respectively… For vjI, where the index corresponding where the index corresponding to item vj is represented as 1 and the rest of the positions are 0.” [*Examiner note: i.e., creating a sparse vector to represent item vj- is being interpreted as processing content data to yield a sparse vector ] [Page 305, 3.1. The Model Architecture – Input Layer])
processing content data comprised of content feature data as input to the content neural network model (Campo: “The plots, the meta data, and the customer data are inputs to the model … The first stage computes the high dimensional embedding of the movie plot (we use movie synopses and long movie scripts) and movie meta data (cast, movie, genre) in a dense space… The final dense representation of the movie plot and associated meta data is computed as the average of all the high-dimensional vectors for all the words and meta data elements for each movie.” [*Examiner note: i.e., the movie plot and meta data are being interpreted as content data, and the final dense representation of the movie plot and associated meta data is being interpreted as a content embedding] [Page 7, 2. System Overview] … “we use a Siamese framework with two identical feed-forward multilayer neural networks” [Page 7, 3. Deep Metric Learning in the Product Space] … “
PNG
media_image6.png
685
747
media_image6.png
Greyscale
” [*Examiner note: (1): the movie plot neural network of the Siamese architecture is being interpreted as a content neural network)] [Page 6, Figure 1]).
Zhang in view of Tareen and Campo both disclose inventions relating to recommending content to a user and are therefore analogous. Zhang in view of Tareen discloses a recommendation system that yields user embeddings and item embeddings. The user embeddings and item embeddings are mapped to a shared dimensional space to yield a content shared-item embedding. Campo discloses a recommendation system that uses a content neural network to generate the item embeddings. It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to generate item embeddings using a content neural network as disclosed by Campo. One would have been motivated to make this change because having a content neural network allows the item vector to place more weight on more relevant aspects of the item (Campo: “In addition to the plots, this step also requires collecting meta data about those movies, like movie genre, casting, etc. … We do this because we want to place as much weight as possible on the plot description and the cast. The plots, the meta data, and the customer data are inputs to the model” [Page 3, 2. System Overview]).
and processing the content embedding through the matchmaking neural network, yielding a content shared-item embedding (Zhang: “each dense representation can be regarded as the latent feature vector for an item. Hence, for an item vj, we formulate the item embedding vector as vj = QT vjI, where Q ∈ ℝK×M, denoting the latent vector matrix for all items. M is the number of items and K is the dimensionality of the user/item latent vector.” [*Examiner note: i.e., the item embedding vector vj is being interpreted as a content shared-item embedding] [Page 306, 3.1. Model Architecture – Adaptive User Representation]);
wherein applying analysis of the user matchmaking embedding in selecting at least one content item comprises: calculating personalization scores between the user shared-item embedding and content shared-item embeddings of a set of candidate content items (Zhang: “for each user ui ∈ U, the recommendation task is defined as recommending a list of items from Ri− that the user ui may be interested in.” [*Examiner note: i.e., the item set Ri− is being interpreted as a set of candidate content items] [Page 304, 2. Problem Definition] … “for an item vj, we formulate the item embedding vector as vj = QT vjI” [*Examiner note: i.e., the item embedding vector vj is being interpreted as a content shared-item embedding] [Page 306, 3.1. Model Architecture – Adaptive User Representation] … “the adaptive user representation uij is derived by weighting M according to the candidate item vj” [*Examiner note: i.e., the adaptive user representation is being interpreted as a user shared-item embedding] [Page 306, 3.1. The Model Architecture – Adaptive User Representation] … “when making predictions, we feed the final adaptive user representation uij and the target item embedding vj into the Euclidean distance: d(i, j) = || uij – vj- ||” [*Examiner note: i.e., Euclidean distance is being interpreted as the calculated personalization score between a user shared-item embedding and a content shared-item embedding] [Page 307, 3.1. Model Architecture – Prediction Layer]);
and updating prioritization of the set of candidate content items based in part on a calculated personalization scores (Zhang: “our model, like CML, also uses rank-based weighting scheme called Weighted Approximate-Rank Pairwise (WARP) loss, which was proposed by Weston et al. [16], to penalize positive items at a lower rank. As a result, we penalize the positive item vj based on its rank by setting σij = log(rankd(i, j) + 1), where rankd(i, j) denotes the rank of item vj in the recommended list of user ui.” [*Examiner note: i.e., the rank-based weighing is being interpreted as updating prioritization of the set of candidate content items] [Page 307, 3.2. Model Inference] … when making predictions, we feed the final adaptive user representation uij and the target item embedding vj into the Euclidean distance: d(i, j) = || uij – vj- ||” [*Examiner note: i.e., the recommended list is based in part on the Euclidean distance (i.e., the calculated personalization scores)] [Page 307, 3.1. Model Architecture – Prediction Layer]).
Regarding claim 18, Zhang discloses:
processing user data comprised of user feature data [as input to a user neural network] and yielding a user embedding (Zhang: “As shown in Fig. 2, the bottom input layer consists of two sparse vectors uiU and vjI, respectively… For uiU, item IDs visited by user ui are represented as 1, and the unvisited items are 0” [*Examiner note: i.e., visited and unvisited items are being interpreted as user feature data, which is being used to represent the user as a vector (i.e., yielding a user embedding)] [Page 305, 3.1. The Model Architecture – Input Layer]);
processing the user embedding through a matchmaking neural network, which is a trained model to map user embeddings and content embeddings to a shared dimensional space, and yielding a user shared-item embedding (Zhang: “We use a memory component and an attention mechanism to flexibly acquire adaptive user representations based on candidate items” [*Examiner note: i.e., the adaptive user representation is being interpreted as a user shared-item embedding] [Page 303, 1. Introduction] … “Hence, the adaptive user representation uij is derived by weighting M according to the candidate item vj as:
PNG
media_image1.png
83
163
media_image1.png
Greyscale
, where … Ri+ denotes historical items liked by user ui” [Page 306, 3.1. The Model Architecture – Adaptive User Representation] … “
PNG
media_image2.png
559
754
media_image2.png
Greyscale
” [*Examiner note: (1): the user embedding uiU and (2): the item embedding vIj are mapped to a shared dimensional space, yielding (3): a user shared-item embedding uij] [Page 305, Fig. 2.]);
and applying analysis of the user shared-item embedding in selecting at least one content item associated with a content shared-item embedding within the matchmaking neural network (Zhang: “Ri− denotes the remaining items. Then, given user set U and item set I, for each user ui ∈ U, the recommendation task is defined as recommending a list of items from Ri− that the user ui may be interested in.” [*Examiner note: i.e., the process of recommending a list of items is being interpreted as applying analysis of the user shared-item embedding] [Page 304, 2. Problem Definition] … “For example, for user #541, the target item #1486 is a positive example in the testset, while items #856, #85, and #237 are historical ones. We can see that AdaCML assigns higher weights on items #856 and #85, a lower weight on item #237, successfully evaluating the target item #1486 as the item desired by the user.” [*Examiner note: i.e., evaluating the target item as the item desired by the user is being interpreted as selecting at least one content item associated with a positive example in the test set (i.e., associated with a content shared-item embedding)] [Page 313, 4.4. The Interpretability of AdaCML, Table 3]).
Zhang fails to disclose:
A system comprising of: one or more computer-readable mediums storing instructions that, when executed by the one or more computer processors, cause a computing platform to perform operations comprising,
processing user data comprised of user feature data as input to a user neural network;
However, Campo discloses:
processing user data comprised of user feature data as input to a user neural network and yielding a user embedding (Campo: “The plots, the meta data, and the customer data are inputs to the model.” [*Examiner note: i.e., the customer data is being interpreted as user feature data] [Page 3, 2. System Overview] … “we calculate, for each customer for each customer, the vector representing the purchase history of that customer” [*Examiner note: the customer history vector is being interpreted as a yielded user embedding] [Page 4, 4. System Implementation] … “we use a Siamese framework with two identical feed-forward multilayer neural networks” [Page 3, 3. Deep Metric Learning in the Product Space] … “
PNG
media_image3.png
782
850
media_image3.png
Greyscale
” [*Examiner note: (1): the customer history neural network of the Siamese framework is being interpreted as a user neural network)] [Page 2, Figure 1. Architecture of the prediction system]);
Zhang and Campo both disclose inventions relating to recommending content to a user and are therefore analogous. Zhang discloses a recommendation system that yields user embeddings and item embeddings. The user embeddings and item embeddings are mapped to a shared dimensional space to yield a user shared-item embedding. Campo discloses a recommendation system that uses a user neural network to generate the user embeddings. It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to generate user embeddings using a user neural network as disclosed by Campo. One would have been motivated to make this change because having a user neural network allows for the user vector to be adapted based on changes in customer preferences (Campo: “we reconstruct customer vectors based on past product purchases and a time discount function [13]. We do this to account for the fact that customers ‘drift’ in the product space because of changes in preferences and we would miss this important mechanism if we employed latent customer vectors” [Page 2, 1. Introduction]).
Zhang in view of Campo fails to disclose:
A system comprising of: one or more computer-readable mediums storing instructions that, when executed by the one or more computer processors, cause a computing platform to perform operations comprising:
However, Tareen discloses:
A system comprising of: one or more computer-readable mediums storing instructions that, when executed by the one or more computer processors, cause a computing platform to perform operations comprising (Tareen: “a computer program product embodied on a computer readable storage medium; and/or a processor, such as a processor configured to execute instructions stored on and/or provided by a memory coupled to the processor.” [Col. 1, lines 55 – 58]):
Zhang in view of Campo and Tareen both disclose inventions relating to recommending content to a user and are therefore analogous. It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to render the teachings of Zhang in view of Campo executable by a computer by storing instructions onto one or more computer-readable mediums using the teachings of Tareen. One having ordinary skill in the art would have been motivated to make this change because computers can be adapted and specialized for specific tasks (Tareen: “Unless stated otherwise, a component such as a processor or a memory described as being configured to perform a task may be implemented as a general component that is temporarily configured 65 to perform the task at a given time or a specific component that is manufactured to perform the task.” [Col. 1, lines 62 – 67]).
Regarding claim 19, Zhang in view of Campo and in further view of Tareen also discloses:
wherein the instructions further cause the computing platform to perform the operations: training the user neural network (Campo: “To prepare the training dataset for the Siamese network, we use the customer purchase data to build the Cartesian product of the purchase history of each customer… To train the Siamese network, we append the fixed-length embedding vectors for each of the two movies in each row. We use the resulting dataset to train the Siamese network” [Page 8, 4. System Implementation] … “
PNG
media_image4.png
520
565
media_image4.png
Greyscale
” [*Examiner note: (1): the customer purchase data is input to the customer history neural network (i.e., the user neural network), therefore (2) training on the customer history trains the user neural network] [Page 6, Figure 1]);
training a content neural network (Campo: “The first stage computes the high dimensional embedding of the movie plot … This first stage is trained on data for hundreds of movies” [Page 3, 2. System Overview] … “
PNG
media_image5.png
520
565
media_image5.png
Greyscale
” [*Examiner note: (1): the first stage handles movie plots, therefore (2) training the first stage trains the movie plot neural network (i.e., the content neural network).] [Page 2, Figure 1])
and training the matchmaking neural network by applying collaborative metric learning (Zhang: “After getting the adaptive user representation, we use the metric learning method to train and optimize our model AdaCML.” [*Examiner note: i.e., applying collaborative metric learning] [Page 305, 3.1. The Model Architecture]).
for a set of content items: processing content data comprised of content feature data [as input to the content neural network model] and yielding a content embedding (Zhang: “As shown in Fig. 2, the bottom input layer consists of two sparse vectors uiU and vjI, that describe user ui and item vj respectively… For vjI, where the index corresponding where the index corresponding to item vj is represented as 1 and the rest of the positions are 0.” [*Examiner note: i.e., creating a sparse vector to represent item vj- is being interpreted as processing content data to yield a sparse vector ] [Page 305, 3.1. The Model Architecture – Input Layer])
processing content data comprised of content feature data as input to the content neural network model (Campo: “The plots, the meta data, and the customer data are inputs to the model … The first stage computes the high dimensional embedding of the movie plot (we use movie synopses and long movie scripts) and movie meta data (cast, movie, genre) in a dense space… The final dense representation of the movie plot and associated meta data is computed as the average of all the high-dimensional vectors for all the words and meta data elements for each movie.” [*Examiner note: i.e., the movie plot and meta data are being interpreted as content data, and the final dense representation of the movie plot and associated meta data is being interpreted as a content embedding] [Page 7, 2. System Overview] … “we use a Siamese framework with two identical feed-forward multilayer neural networks” [Page 7, 3. Deep Metric Learning in the Product Space] … “
PNG
media_image6.png
685
747
media_image6.png
Greyscale
” [*Examiner note: (1): the movie plot neural network of the Siamese architecture is being interpreted as a content neural network)] [Page 6, Figure 1]).
Zhang in view of Tareen and Campo both disclose inventions relating to recommending content to a user and are therefore analogous. Zhang in view of Tareen discloses a recommendation system that yields user embeddings and item embeddings. The user embeddings and item embeddings are mapped to a shared dimensional space to yield a content shared-item embedding. Campo discloses a recommendation system that uses a content neural network to generate the item embeddings. It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to generate item embeddings using a content neural network as disclosed by Campo. One would have been motivated to make this change because having a content neural network allows the item vector to place more weight on more relevant aspects of the item (Campo: “In addition to the plots, this step also requires collecting meta data about those movies, like movie genre, casting, etc. … We do this because we want to place as much weight as possible on the plot description and the cast. The plots, the meta data, and the customer data are inputs to the model” [Page 3, 2. System Overview]).
and processing the content embedding through the matchmaking neural network, yielding a content shared-item embedding (Zhang: “each dense representation can be regarded as the latent feature vector for an item. Hence, for an item vj, we formulate the item embedding vector as vj = QT vjI, where Q ∈ ℝK×M, denoting the latent vector matrix for all items. M is the number of items and K is the dimensionality of the user/item latent vector.” [*Examiner note: i.e., the item embedding vector vj is being interpreted as a content shared-item embedding] [Page 306, 3.1. Model Architecture – Adaptive User Representation]);
wherein applying analysis of the user matchmaking embedding in selecting at least one content item comprises: calculating personalization scores between the user shared-item embedding and content shared-item embeddings of a set of candidate content items (Zhang: “for each user ui ∈ U, the recommendation task is defined as recommending a list of items from Ri− that the user ui may be interested in.” [*Examiner note: i.e., the item set Ri− is being interpreted as a set of candidate content items] [Page 304, 2. Problem Definition] … “for an item vj, we formulate the item embedding vector as vj = QT vjI” [*Examiner note: i.e., the item embedding vector vj is being interpreted as a content shared-item embedding] [Page 306, 3.1. Model Architecture – Adaptive User Representation] … “the adaptive user representation uij is derived by weighting M according to the candidate item vj” [*Examiner note: i.e., the adaptive user representation is being interpreted as a user shared-item embedding] [Page 306, 3.1. The Model Architecture – Adaptive User Representation] … “when making predictions, we feed the final adaptive user representation uij and the target item embedding vj into the Euclidean distance: d(i, j) = || uij – vj- ||” [*Examiner note: i.e., Euclidean distance is being interpreted as the calculated personalization score between a user shared-item embedding and a content shared-item embedding] [Page 307, 3.1. Model Architecture – Prediction Layer]);
and updating prioritization of the set of candidate content items based in part on a calculated personalization scores (Zhang: “our model, like CML, also uses rank-based weighting scheme called Weighted Approximate-Rank Pairwise (WARP) loss, which was proposed by Weston et al. [16], to penalize positive items at a lower rank. As a result, we penalize the positive item vj based on its rank by setting σij = log(rankd(i, j) + 1), where rankd(i, j) denotes the rank of item vj in the recommended list of user ui.” [*Examiner note: i.e., updating the prioritization based on the pairs rank in the predicted list. The Euclidean distance was used to predict the list (i.e., based in part on the calculated personalization scores] [Page 307, 3.2. Model Inference]).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Brent Hoover whose telephone number is (303)297-4403. The examiner can normally be reached Monday - Friday 9-5 MST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Abdullah Kawsar can be reached at 571-270-3169. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/BRENT JOHNSTON HOOVER/Primary Examiner, Art Unit 2127
Prosecution Insights