DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
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.
Claims 5–9 and 11–20 are rejected under 35 U.S.C. 112(b) as being indefinite for failing to particularly point out and distinctly claim the subject matter that the inventor or a joint inventor regards as the invention.
Regarding claim 5, the claim recites “wherein concatenating the respective embeddings vectors to determine the respective pairwise similarities among the embeddings vectors comprising” applying the first embeddings vector index as keys and values and applying the second embeddings vector index as queries. It is unclear how assigning the respective indexes as keys, values, and queries constitutes or limits the recited concatenation. The claim does not identify what vectors or indexes are concatenated as part of these operations. Accordingly, the scope of the recited “concatenating” cannot be determined.
Regarding claims 6, 14, and 19, each claim introduces “a respective target embeddings vector” corresponding to a weighted average of the attention embeddings vectors but subsequently recites that the predicted next user action is determined based on “target item embeddings vectors.” Claims 7, 8, 9, 15, and 20 then refer to “the target embeddings vectors.” It is unclear whether “the target embeddings vectors,” “target item embeddings vectors,” and “a respective target embeddings vector” identify the same element or different elements. It is also unclear whether the claims require one target vector or multiple target vectors. Claims 7–9, 15, and 20 are further indefinite by virtue of their dependency or reference to this unclear terminology.
Regarding claims 7, 15, and 20, each claim recites generating “a respective candidate embeddings vector” based on a plurality of predicted user actions but subsequently refers to plural “candidate embeddings vectors.” The claims do not establish the correspondence denoted by “respective.” It is unclear whether one candidate embeddings vector is generated for each predicted user action, whether multiple candidate embeddings vectors are generated for each predicted user action, or whether a single candidate embeddings vector represents the plurality collectively. Claim 7 further recites determining “a respective similarity” and then determines the predicted next user action based on plural “similarities,” without identifying the correspondence between the candidate embeddings vectors, target embeddings vectors, and similarities. Claims 8 and 9 inherit this uncertainty.
Regarding claims 8, 15, and 20, the limitation requiring “applying the respective similarity score to the respective target embeddings vector” does not identify what operation is performed on the target embeddings vector. It is unclear whether the similarity score weights, scales, ranks, selects, modifies, or otherwise operates on the target embeddings vector. These claims also determine the similarity score “based on applying a SoftMax loss function” without identifying whether the function is applied to the candidate embeddings vectors, the target embeddings vectors, dot products, logits, or previously determined similarity values. Accordingly, the metes and bounds of these limitations cannot be determined.
Regarding independent claim 11, the claim recites receiving “a first sequence of user actions” and “a second sequence of user actions,” but subsequently recites outputting the predicted next user action “to the user in response to receiving the sequence.” The terms “the user” and “the sequence” lack proper antecedent basis. The claim does not introduce a user, establish that both sequences concern the same user, or identify whether “the sequence” refers to the first sequence, the second sequence, or a combination of the two sequences. Claims 12–16 are likewise indefinite by virtue of their dependency from claim 11.
Independent claim 17 contains the same defects. Claim 17 introduces a first sequence of user actions and a second sequence of user actions but subsequently requires outputting the predicted next user action “to the user in response to receiving the sequence.” The claim does not introduce “the user,” establish that the two sequences concern the same user, or identify which sequence constitutes “the sequence.” Claims 18–20 are likewise indefinite by virtue of their dependency from claim 17.
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 1-20 are rejected under 35 U.S.C. 101 because they are directed to an abstract idea without significantly more.
Regarding claim 1:
Step 1: is the claim directed to one of the four statutory categories?
Yes, the claim is directed to a method.
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
Yes, the limitations: “generating a respective embeddings vector representative of each of the user actions;” and “generating a respective attention embeddings vector for each of the user actions according to the similarities;” are directed to a mathematical concept merely implemented on a computer under MPEP 2106.04(a)(2)(I).
Further, the limitations: “determining respective pairwise similarities among the embeddings vectors;” is directed to a mental process of evaluation under MPEP 2106.04(a)(2)(III).
Step 2A, prong 2: Do the additional elements integrate into a practical application?
No. The limitations: “receiving a sequence of user actions for a user;” and “and outputting the predicted next user action to the user in response to receiving the sequence” are directed to mere data gathering and outputting under MPEP 2106.05(g).
Further, the limitation: “applying a predictive model to the attention embeddings vectors to determine a predicted next user action;” is directed to mere instructions to apply under MPEP 2106.05(f).
Step 2B: Does the claim recite additional elements that amount to significantly more than the
judicial exception?
No. The limitations: “receiving a sequence of user actions for a user;” and “and outputting the predicted next user action to the user in response to receiving the sequence” are directed to the well-understood, routine, and conventional activity of “Receiving or transmitting data over a network” under MPEP 2106.05(d).
Further, the limitation: “applying a predictive model to the attention embeddings vectors to determine a predicted next user action;” is directed to mere instructions to apply under MPEP 2106.05(f).
Regarding claim 2:
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
Yes, the claim is dependent on claim 1.
Step 2A, prong 2: Do the additional elements integrate into a practical application?
No. The limitations: “receiving a first sequence of user actions based on a first parameter; and receiving a second sequence of user actions based on a second parameter” are directed to mere data gathering and outputting under MPEP 2106.05(g).
Step 2B: Does the claim recite additional elements that amount to significantly more than the
judicial exception?
No. The limitations: “receiving a first sequence of user actions based on a first parameter; and receiving a second sequence of user actions based on a second parameter” are directed to the well-understood, routine, and conventional activity of “Receiving or transmitting data over a network” under MPEP 2106.05(d).
Regarding claim 3:
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
Yes. The limitations: “generating a first embeddings vector index based on the first sequence of user actions;” and “and generating a second embeddings vector index based on the second sequence of user actions;” are directed to a mathematical concept merely implemented on a computer under MPEP 2106.04(a)(2)(I).
Further, the limitation: “wherein the respective pairwise similarities are determined based on respective embeddings vectors from the first embeddings vector index and the second embeddings vector index” is directed to a mental process of evaluation under MPEP 2106.04(a)(2)(III).
Regarding claim 4:
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
Yes. The limitation: “applying a causal mask to the respective embeddings vector for each of the user actions; and concatenating the respective embeddings vectors to determine the respective pairwise similarities among the embeddings vectors” is directed to a mathematical concept merely implemented on a computer under MPEP 2106.04(a)(2)(I).
Examiner notes that concatenation in machine learning is “the appending of vectors or matrices to form a new vector or matrix.” See attached NPL: The Science of Machine Learning & AI.
Regarding claim 5:
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
Yes. The limitation: “wherein the attention embeddings vectors are generated based on the first embeddings vector index and the second embeddings vector index” is directed to a mental process of evaluation under MPEP 2106.04(a)(2)(III).
Step 2A, prong 2: Do the additional elements integrate into a practical application?
No. The limitation: “applying the first embeddings vector index as keys and values, and applying the second embeddings vector index as queries,” are directed to field of use under MPEP 2106.05(h).
Step 2B: Does the claim recite additional elements that amount to significantly more than the
judicial exception?
No. The limitation: “applying the first embeddings vector index as keys and values, and applying the second embeddings vector index as queries,” are directed to field of use under MPEP 2106.05(h).
Regarding claim 6:
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
Yes. The limitation: “determining a respective target embeddings vector corresponding to a weighted average of the attention embeddings vectors;” is directed to a mathematical concept merely implemented on a computer under MPEP 2106.04(a)(2)(I).
Further, the limitation: “wherein the predicted next user action is determined based on target item embeddings vectors” is directed to a mental process of evaluation under MPEP 2106.04(a)(2)(III).
Regarding claim 7:
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
Yes. The limitations: “determining a plurality of predicted user actions based on the attention embeddings vectors;” and “and determining a respective similarity between the candidate embeddings vectors and the target embeddings vectors; wherein the predicted next user action is determined based on the similarities between the target embeddings vectors and the candidate embeddings vectors” are directed to a mental process of judgment under MPEP 2106.04(a)(2)(III).
Further, the limitation: “generating a respective candidate embeddings vector based on the plurality of predicted user actions;” is directed to a mathematical concept merely implemented on a computer under MPEP 2106.04(a)(2)(I).
Regarding claim 8:
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
Yes. The limitation: “determining a respective similarity score between the target embeddings vectors and the candidate embeddings vectors based on applying a SoftMax loss function;” is directed to a mathematical concept merely implemented on a computer under MPEP 2106.04(a)(2)(I).
Step 2A, prong 2: Do the additional elements integrate into a practical application?
No. The limitation: “and applying the respective similarity score to the respective target embeddings vector” is directed to mere data gathering under MPEP 2106.05(g).
Step 2B: Does the claim recite additional elements that amount to significantly more than the
judicial exception?
No. The limitation: “and applying the respective similarity score to the respective target embeddings vector” is directed to the well-understood, routine, and conventional activity of “Receiving or transmitting data over a network” under MPEP 2106.05(d).
Regarding claim 9:
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
Yes. The limitation: “wherein the predicted next user action is determined based on the respective similarity score between the target embeddings vectors and the candidate embeddings vectors” is directed to a mental process of judgment under MPEP 2106.04(a)(2)(III).
Regarding claim 10:
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
Yes. The limitation is dependent on claim 1.
Step 2A, prong 2: Do the additional elements integrate into a practical application?
No. The limitation: “wherein the embeddings vectors are representative of metadata generated based on the user actions” is directed to field of use under MPEP 2106.05(h).
Step 2B: Does the claim recite additional elements that amount to significantly more than the
judicial exception?
No. The limitation: “wherein the embeddings vectors are representative of metadata generated based on the user actions” is directed to field of use under MPEP 2106.05(h).
Regarding claim 11:
Step 1: is the claim directed to one of the four statutory categories?
Yes, the claim is directed to a machine.
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
Yes. The limitation: “generate a respective embeddings vector representative of each of the user actions for the first sequence of user actions and the second sequence of user actions;” and “generate a respective attention embeddings vector for each of the user actions according to the similarities;” are directed to a mathematical concept under MPEP 2106.04(a)(2)(I).
Further, the limitation: “determine respective pairwise similarities among the embeddings vectors;” is directed to a mental process of evaluation under MPEP 2106.04(a)(2)(III).
Step 2A, prong 2: Do the additional elements integrate into a practical application?
No. The limitations: “a processor; and a non-transitory computer readable media having stored therein instructions that are executable by the processor to perform operations comprising: receive a first sequence of user actions based on a first parameter;” and “apply a predictive model to the attention embeddings vectors to determine a predicted next user action;” are directed to mere instructions to apply an exception under MPEP 2106.05(f).
Further, the limitations: “receive a first sequence of user actions based on a first parameter;” “receive a second sequence of user actions based on a second parameter;” and “and output the predicted next user action to the user in response to receiving the sequence” are directed to mere data gathering and outputting under MPEP 2106.05(g).
Step 2B: Does the claim recite additional elements that amount to significantly more than the
judicial exception?
No. The limitations: “a processor; and a non-transitory computer readable media having stored therein instructions that are executable by the processor to perform operations comprising: receive a first sequence of user actions based on a first parameter;” are directed to generic computing components under MPEP 2106.05(f).
Further, the limitations: “receive a first sequence of user actions based on a first parameter;” and “receive a second sequence of user actions based on a second parameter;” and “and output the predicted next user action to the user in response to receiving the sequence” are directed to the well-understood, routine, and conventional activity of “Receiving or transmitting data over a network” under MPEP 2106.05(d).
Further, the limitation: “apply a predictive model to the attention embeddings vectors to determine a predicted next user action;” are directed to mere instructions to apply an exception under MPEP 2106.05(f).
Regarding claim 12:
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
Yes. The limitations: “generate a first embeddings vector index based on the first sequence of user actions; and generate a second embeddings vector index based on the second sequence of user actions;” are directed to a mathematical concept merely implemented on a computer under MPEP 2106.04(a)(2)(I).
Further, the limitation: “wherein the respective pairwise similarities are determined based on respective embeddings vectors from the first embeddings vector index and the second embeddings vector index” is directed to a mental process of evaluation under MPEP 2106.04(a)(2)(III).
Regarding claim 13:
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
Yes. The limitations: “apply a causal mask to the respective embeddings vector for each of the user actions;” and “and concatenate the first embeddings vector index and the second embeddings vector index to determine the respective pairwise similarities among the embeddings vectors; wherein the attention embeddings vectors are generated based on the first embeddings vector index and the second embeddings vector index” are directed to a mathematical concept merely implemented on a computer under MPEP 2106.04(a)(2)(I).
Step 2A, prong 2: Do the additional elements integrate into a practical application?
No. The limitations: “apply the first embeddings vector index as keys and values; apply the second embeddings vector index as queries;” are directed to field of use under MPEP 2106.05(h).
Step 2B: Does the claim recite additional elements that amount to significantly more than the
judicial exception?
No. The limitations: “apply the first embeddings vector index as keys and values; apply the second embeddings vector index as queries;” are directed to field of use under MPEP 2106.05(h).
Regarding claim 14:
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
Yes. The limitations: “determine a respective target embeddings vector corresponding to a weighted average of the attention embeddings vectors;” is directed to a mathematical concept merely implemented on a computer under MPEP 2106.04(a)(2)(I).
Further, “wherein the predicted next user action is determined based on target item embeddings vectors” is directed to a mental process of evaluation under MPEP 2106.04(a)(2)(III).
Regarding claim 15:
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
No. The limitations: “determine a plurality of predicted user actions based on the attention embeddings vectors;” and “wherein the predicted next user action is determined based on the respective similarity score between the target embeddings vectors and the candidate embeddings vectors” are directed to a mental process of evaluation under MPEP 2106.04(a)(2)(III).
Further, the limitations: “generate a respective candidate embeddings vector based on the plurality of predicted user actions; determine a respective similarity score between the target embeddings vectors and the candidate embeddings vectors based on applying a SoftMax loss function; and apply the respective similarity score to the respective target embeddings vector;” are directed to mathematical concepts merely implemented on a computer under MPEP 2106.04(a)(2)(I).
Regarding claim 16:
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
Yes. The limitation is dependent on claim 11.
Step 2A, prong 2: Do the additional elements integrate into a practical application?
No. The limitation: “wherein the embeddings vectors are representative of metadata generated based on the first sequence of user actions and the second sequence of user actions” is directed to field of use under MPEP 2106.05(h).
Step 2B: Does the claim recite additional elements that amount to significantly more than the
judicial exception?
No. The limitation: “wherein the embeddings vectors are representative of metadata generated based on the first sequence of user actions and the second sequence of user actions” is directed to field of use under MPEP 2106.05(h).
Regarding claim 17:
Step 1: is the claim directed to one of the four statutory categories?
Yes. The limitation is directed to a machine.
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
Yes. The limitations: “generate a respective embeddings vector for a first embeddings vector index based on the first sequence of user actions; receive a second sequence of user actions based on a second parameter; generate a respective embeddings vector for a second embeddings vector index based on the second sequence of user actions;” are directed to mathematical concepts under MPEP 2106.04(a)(2)(I).
Further, the limitations “determine respective pairwise similarities among the embeddings vectors” and “generate a respective attention embeddings vector for each of the user actions according to the similarities” are directed to a mental process of evaluation under MPEP 2106.04(a)(2)(III).
Step 2A, prong 2: Do the additional elements integrate into a practical application?
No. The limitations: “receive a first sequence of user actions based on a first parameter;” and “wherein the respective pairwise similarities are determined based on respective embeddings vectors from the first embeddings vector index and the second embeddings vector index” and “output the predicted next user action to the user in response to receiving the sequence;” are directed to mere data gathering under MPEP 2106.05(g).
Further, the limitation: “wherein the embeddings vectors are representative of metadata generated based on the user actions;” is directed to field of use under MPEP 2106.05(h).
Step 2B: Does the claim recite additional elements that amount to significantly more than the
judicial exception?
No. The limitations: “receive a first sequence of user actions based on a first parameter;” and “wherein the respective pairwise similarities are determined based on respective embeddings vectors from the first embeddings vector index and the second embeddings vector index” and “output the predicted next user action to the user in response to receiving the sequence;” are directed to the well-understood, routine, and conventional activity of “Receiving or transmitting data over a network” under MPEP 2106.05(d).
Further, the limitation: “wherein the embeddings vectors are representative of metadata generated based on the user actions;” is directed to field of use under MPEP 2106.05(h).
Claim 18 is rejected under the same rationale as claim 13.
Regarding claim 19:
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
No. The limitation: “determine a respective target embeddings vector corresponding to a weighted average of the attention embeddings vectors;” is directed to a mathematical concept under MPEP 2106.04(a)(2)(I).
Further, the limitation: “wherein the predicted next user action is determined based on target item embeddings vectors” is directed to a mental process of evaluation under MPEP 2106.04(a)(2)(III).
Regarding claim 20:
Step 2A, prong 1: Is the claim directed to a law of nature, a natural phenomenon, or an abstract
idea?
Yes. The limitation: “determine a plurality of predicted user actions based on the attention embeddings vectors;” and “generate a respective candidate embeddings vector based on the plurality of predicted user actions;” are directed to a mental process of evaluation under MPEP 2106.04(a)(2)(III).
Further, the limitations: “determining a respective similarity score between the target embeddings vectors and the candidate embeddings vectors based on applying a SoftMax loss function; and apply the respective similarity score to the respective target embeddings vector; wherein the predicted next user action is determined based on the respective similarity score between the target embeddings vectors and the candidate embeddings vectors” are directed to mathematical concept merely implemented on a computer under MPEP 2106.04(a)(2)(I).
Claim Rejections - 35 USC § 103
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, 6, 10-11, 14, 16-17, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over US Pre-Grant Patent 2022/0107852 (Kulkarni et al; Kulkarni) in view of “Self-Attentive Sequential Recommendation,” arXiv, Kang et al; Kang, 2018.
Regarding claim 1:
Kulkarni teaches:
1. A method for predicting a next user action at an electronic user interface, the method comprising:
(Kulkarni, ¶0007)
“Embodiments of the present disclosure provide benefits and/or solve one or more of the foregoing or other problems in the art with systems, non-transitory computer-readable media, and methods that utilize a natural language model to predict a next user activity event based on an ordered series of preceding user activity events [i.e. A method for predicting a next user action at an electronic user interface, the method comprising:].”
2. receiving a sequence of user actions for a user;
(Kulkarni, ¶0140)
“In yet another example an additional act not shown in FIG. 8, act(s) in the series of acts 800 may include an act to: identify, from an events database, a sequence of activity events, each activity event within the sequence of activity events comprising raw event data associated with a user account of a content management system [i.e. receiving a sequence of user actions for a user;].”
3. generating a respective embeddings vector representative of each of the user actions;
(Kulkarni, ¶0021)
“The user activity sequence system can then convert (e.g., transform) these or other features for a given user activity event to vector embeddings [i.e. generating a respective embeddings vector representative of each of the user actions;].”
4. and outputting the predicted next user action to the user in response to receiving the sequence.
(Kulkarni, ¶0007)
“In turn, the disclosed systems can select, as the predicted next user activity event, the most probable next user activity event represented in a candidate sequence of activity events. In accordance with the selected next user activity event…[i.e. and outputting the predicted next user action to the user in response to receiving the sequence.]”
Kang teaches:
1. determining respective pairwise similarities among the embeddings vectors;
(Kang, pg. 3, Sect. III.B, col.2, ¶1, Eq. 2)
“The scaled dot-product attention [3] is defined as
PNG
media_image1.png
33
310
media_image1.png
Greyscale
where Q represents the queries, K the keys and V the values (each row represents an item). Intuitively, the attention layer calculates a weighted sum of all values, where the weight between query i and value j relates to the interaction between query i and key j [i.e. determining respective pairwise similarities among the embeddings vectors;].”
2. generating a respective attention embeddings vector for each of the user actions according to the similarities;
(Kang, pg. 3, Sect. III.B, col.2, ¶2, Eq. 3)
“Recently, a self-attention method was proposed which uses the same objects as queries, keys, and values [3]. In our case, the self-attention operation takes the embedding E as input, converts it to three matrices through linear projections, and feeds them into an attention layer: [i.e. generating a respective attention embeddings vector for each of the user actions according to the similarities;]”
PNG
media_image2.png
24
333
media_image2.png
Greyscale
3. applying a predictive model to the attention embeddings vectors to determine a predicted next user action;
(Kang, pg. 3, Sect. III.B, col.2, ¶3)
“Due to the nature of sequences, the model should consider only the first t items when predicting the (t + 1)-st item [i.e. to determine a predicted next user action;].”
(Kang, pg. 3, Sect. III.B, col.2, ¶4, Eq. 4)
“To endow the model with nonlinearity and to consider interactions between different latent dimensions, we apply a point-wise two-layer feed-forward network to all Si identically (sharing parameters):
PNG
media_image3.png
25
352
media_image3.png
Greyscale
where W(1),W(2) are d × d matrices and b(1),b(2) are d dimensional vectors. Note that there is no interaction between Si and Sj (i= j), meaning that we still prevent information leaks (from back to front) [i.e. applying a predictive model to the attention embeddings vectors].”
One of ordinary skill, at the time the invention was filed, would have been motivated to modify Kulkarni with Kang. It would have been obvious to implement Kulkarni’s predictive model using Kang’s per-action self-attention and prediction layers. Kulkarni expressly contemplates transformer layers and neural-attention mechanisms in ¶[0043], and ¶[0095] permits directly supplying activity-event vectors to the neural network. Kang supplies the implementation for weighting relevant preceding actions and using the resulting representations to predict the next action. The combination retains Kulkarni’s sequence input and user-facing output while implementing its predictive model with Kang’s disclosed attention architecture “to identify which items are ‘relevant’ from a user’s action history, and use them to predict the next item (Kang, Abstract).”
Regarding claim 2:
Kulkarni and Kang teach the method of claim 1.
Kulkarni teaches:
1. receiving a first sequence of user actions based on a first parameter;
(Kulkarni, ¶0083)
“In at least one embodiment, the user activity sequence system 104 constructs the series of sequential tokens at act 312 to form a sentence [i.e. receiving a first sequence of user actions based on a first parameter;].”
2. and receiving a second sequence of user actions based on a second parameter.
(Kulkarni, ¶0085)
“In some cases, generating paragraphs or documents comprising multiple sentences of sequential tokens allows the user activity sequence system 104 to provide more context when predicting a next activity event [i.e. and receiving a second sequence of user actions based on a second parameter].”
One of ordinary skill, at the time the invention was filed, would have been motivated to modify Kulkarni with Kang. The motivation is the same as claim 1.
Regarding claim 3:
Kulkarni and Kang teach the method of claim 1.
Kang teaches:
1. generating a first embeddings vector index based on the first sequence of user actions; and generating a second embeddings vector index based on the second sequence of user actions;
(Kang, pg. 3, Sect.IIIA, col. 1, ¶3)
“As we will see in the next section, since the self-attention model doesn’t include any recurrent or convolutional module, it is not aware of the positions of previous items. Hence we inject a learnable position embedding P∈Rn×d into the input embedding…[i.e. generating a first embeddings vector index based on the first sequence of user actions; and generating a second embeddings vector index based on the second sequence of user actions;]”
PNG
media_image4.png
81
346
media_image4.png
Greyscale
3. wherein the respective pairwise similarities are determined based on respective embeddings vectors from the first embeddings vector index and the second embeddings vector index.
(Kang, pg. 3, Sect. III.B, col.2, ¶2, Eq. 3)
“In our case, the self-attention operation takes the embedding E as input, converts it to three matrices through linear projections, and feeds them into an attention layer: [i.e. wherein the respective pairwise similarities are determined based on respective embeddings vectors from the first embeddings vector index and the second embeddings vector index]”
PNG
media_image2.png
24
333
media_image2.png
Greyscale
One of ordinary skill, at the time the invention was filed, would have been motivated to modify Kulkarni with Kang. The motivation is the same as claim 1.
Regarding claim 4:
Kulkarni and Kang teach the method of claim 1.
Kang teaches:
1. applying a causal mask to the respective embeddings vector for each of the user actions;
(Kang, pg. 3, Sect. III.B, col.2, ¶3)
“However, the t-th output of the self-attention layer (St) contains embeddings of subsequent items, which makes the model ill-posed. Hence, we modify the attention by forbidding all links between Qi and Kj (j > i) [i.e. applying a causal mask to the respective embeddings vector for each of the user actions;].”
2. and concatenating the respective embeddings vectors to determine the respective pairwise similarities among the embeddings vectors.
(Kang, pg. 3, Sect.IIIA, col. 1, ¶3)
“Hence we inject a learnable position embedding P∈Rn×d into the input embedding: [i.e. and concatenating the respective embeddings vectors to determine the respective pairwise similarities among the embeddings vectors].
PNG
media_image4.png
81
346
media_image4.png
Greyscale
Examiner notes that Eq. 1 stacks individual embedding vectors into the input matrix, which fulfills the concatenation requirement. See attached NPL: Concatenation — The Science of Machine Learning & AI.
One of ordinary skill, at the time the invention was filed, would have been motivated to modify Kulkarni with Kang. The motivation is the same as claim 1.
Regarding claim 6:
Kulkarni and Kang teach the method of claim 1.
Kulkarni teaches:
1. determining a respective target embeddings vector corresponding to a weighted average of the attention embeddings vectors;
(Kulkarni, ¶0112)
“In this approach, the user activity sequence system 104 can express the order in a weighted sum of the average final embedding layer described in the first approach [i.e. determining a respective target embeddings vector corresponding to a weighted average of the attention embeddings vectors;].”
2. wherein the predicted next user action is determined based on target item embeddings vectors.
(Kulkarni, ¶0112)
“For example, the user activity sequence system 104 may, for each activity event in a sequence of activity events, determine an activity event vector of length 16 [i.e. wherein the predicted next user action is determined based on target item embeddings vectors].”
One of ordinary skill, at the time the invention was filed, would have been motivated to modify Kulkarni with Kang. The motivation is the same as claim 1.
Regarding claim 10:
Kulkarni and Kang teach the method of claim 1.
Kulkarni teaches:
1. wherein the embeddings vectors are representative of metadata generated based on the user actions.
(Kulkarni, ¶0020)
“Specifically, the user activity sequence system can create an activity event vector (e.g., a feature vector) for each user activity event that includes various attributes of a given activity event, for example, a timestamp, action type, device type, filename, and other file metadata [i.e. wherein the embeddings vectors are representative of metadata generated based on the user actions].”
One of ordinary skill, at the time the invention was filed, would have been motivated to modify Kulkarni with Kang. The motivation is the same as claim 1.
Regarding claim 11 and analogous claim 17:
Kulkarni teaches:
1. A system comprising: a processor; and a non-transitory computer readable media having stored therein instructions that are executable by the processor to perform operations comprising:
(Kulkarni, ¶0128)
“For example, the components of the computing device 700 can include one or more instructions stored on a computer-readable storage medium and executable by processors of one or more computing devices, such as a client device or server device. When executed by the one or more processors, the computer-executable instructions of the user activity sequence system 104 can cause the computing device(s) (e.g., the computing device 700) to perform the methods described herein [i.e. A system comprising: a processor; and a non-transitory computer readable media having stored therein instructions that are executable by the processor to perform operations comprising:].”
2. receive a first sequence of user actions based on a first parameter;
(Kulkarni, ¶0083)
“In at least one embodiment, the user activity sequence system 104 constructs the series of sequential tokens at act 312 to form a sentence. For example, the user activity sequence system 104 may construct the series of sequential tokens to comprise the tokens that correspond to activity events occurring within a predetermined time period or between periods of inactivity (e.g., a threshold timeout period, such as thirty minutes, sixty minutes, two hours, five hours, etc.) [i.e. receive a first sequence of user actions based on a first parameter;].”
3. receive a second sequence of user actions based on a second parameter;
(Kulkarni, ¶0084)
“Additionally or alternatively, the user activity sequence system 104 may generate paragraphs or documents of sequential tokens (e.g., comprising multiple sentences of tokens) [i.e. receive a second sequence of user actions based on a second parameter;].”
4. generate a respective embeddings vector representative of each of the user actions for the first sequence of user actions and the second sequence of user actions;
(Kulkarni, ¶0020)
“Specifically, the user activity sequence system can create an activity event vector (e.g., a feature vector) for each user activity event that includes various attributes of a given activity event, for example, a timestamp, action type, device type, filename, and other file metadata. The user activity sequence system can hash the activity event vector to create an event token representing the user activity event [i.e. generate a respective embeddings vector representative of each of the user actions for the first sequence of user actions and the second sequence of user actions;].”
5. and output the predicted next user action to the user in response to receiving the sequence.
(Kulkarni, ¶0007)
“In turn, the disclosed systems can select, as the predicted next user activity event, the most probable next user activity event represented in a candidate sequence of activity events. In accordance with the selected next user activity event…[i.e. and output the predicted next user action to the user in response to receiving the sequence].”
Kang teaches:
1. determine respective pairwise similarities among the embeddings vectors;
(Kang, pg. 3, Sect. III.B, col.2, ¶1, Eq. 2)
“The scaled dot-product attention [3] is defined as
PNG
media_image1.png
33
310
media_image1.png
Greyscale
where Q represents the queries, K the keys and V the values (each row represents an item). Intuitively, the attention layer calculates a weighted sum of all values, where the weight between query i and value j relates to the interaction between query i and key j [i.e. determining respective pairwise similarities among the embeddings vectors;].”
2. generate a respective attention embeddings vector for each of the user actions according to the similarities;
(Kang, pg. 3, Sect. III.B, col.2, ¶2, Eq. 3)
“Recently, a self-attention method was proposed which uses the same objects as queries, keys, and values [3]. In our case, the self-attention operation takes the embedding E as input, converts it to three matrices through linear projections, and feeds them into an attention layer: [i.e. generating a respective attention embeddings vector for each of the user actions according to the similarities;]”
PNG
media_image2.png
24
333
media_image2.png
Greyscale
3. applying a predictive model to the attention embeddings vectors to determine a predicted next user action;
(Kang, pg. 3, Sect. III.B, col.2, ¶3)
“Due to the nature of sequences, the model should consider only the first t items when predicting the (t + 1)-st item [i.e. to determine a predicted next user action;].”
(Kang, pg. 3, Sect. III.B, col.2, ¶4, Eq. 4)
“To endow the model with nonlinearity and to consider interactions between different latent dimensions, we apply a point-wise two-layer feed-forward network to all Si identically (sharing parameters):
PNG
media_image3.png
25
352
media_image3.png
Greyscale
where W(1),W(2) are d × d matrices and b(1),b(2) are d dimensional vectors. Note that there is no interaction between Si and Sj (i= j), meaning that we still prevent information leaks (from back to front) [i.e. applying a predictive model to the attention embeddings vectors].”
One of ordinary skill, at the time the invention was filed, would have been motivated to modify Kulkarni with Kang. The motivation is the same as claim 1.
Regarding claim 12:
Kulkarni and Kang teach the system of claim 11.
Kulkarni teaches:
1. generate a first embeddings vector index based on the first sequence of user actions; and generate a second embeddings vector index based on the second sequence of user actions;
(Kang, pg. 3, Sect.IIIA, col. 1, ¶3)
“As we will see in the next section, since the self-attention model doesn’t include any recurrent or convolutional module, it is not aware of the positions of previous items. Hence we inject a learnable position embedding P∈Rn×d into the input embedding…[i.e. generating a first embeddings vector index based on the first sequence of user actions; and generating a second embeddings vector index based on the second sequence of user actions;]”
PNG
media_image4.png
81
346
media_image4.png
Greyscale
2. wherein the respective pairwise similarities are determined based on respective embeddings vectors from the first embeddings vector index and the second embeddings vector index.
(Kang, pg. 3, Sect. III.B, col.2, ¶2, Eq. 3)
“In our case, the self-attention operation takes the embedding E as input, converts it to three matrices through linear projections, and feeds them into an attention layer: [i.e. wherein the respective pairwise similarities are determined based on respective embeddings vectors from the first embeddings vector index and the second embeddings vector index]”
PNG
media_image2.png
24
333
media_image2.png
Greyscale
One of ordinary skill, at the time the invention was filed, would have been motivated to modify Kulkarni with Kang. The motivation is the same as claim 1.
Regarding claim 14:
Kulkarni and Kang teach the system of claim 11.
Kulkarni teaches:
1. determine a respective target embeddings vector corresponding to a weighted average of the attention embeddings vectors;
(Kulkarni, ¶0112)
“In this approach, the user activity sequence system 104 can express the order in a weighted sum of the average final embedding layer described in the first approach [i.e. determine a respective target embeddings vector corresponding to a weighted average of the attention embeddings vectors;].”
2. wherein the predicted next user action is determined based on target item embeddings vectors.
(Kulkarni, ¶0112)
“For example, the user activity sequence system 104 may, for each activity event in a sequence of activity events, determine an activity event vector of length 16 [i.e. wherein the predicted next user action is determined based on target item embeddings vectors].”
One of ordinary skill, at the time the invention was filed, would have been motivated to modify Kulkarni with Kang. The motivation is the same as claim 1.
Regarding claim 16:
Kulkarni and Kang teach the system of claim 11.
Kulkarni teaches:
1. wherein the embeddings vectors are representative of metadata generated based on the first sequence of user actions and the second sequence of user actions.
(Kulkarni, ¶0020)
“Specifically, the user activity sequence system can create an activity event vector (e.g., a feature vector) for each user activity event that includes various attributes of a given activity event, for example, a timestamp, action type, device type, filename, and other file metadata [i.e. wherein the embeddings vectors are representative of metadata generated based on the first sequence of user actions and the second sequence of user actions].”
One of ordinary skill, at the time the invention was filed, would have been motivated to modify Kulkarni with Kang. The motivation is the same as claim 1.
Regarding claim 19:
Kulkarni and Kang teach the machine of claim 17.
Kulkarni teaches:
1. determine a respective target embeddings vector corresponding to a weighted average of the attention embeddings vectors;
(Kulkarni, ¶0112)
“In this approach, the user activity sequence system 104 can express the order in a weighted sum of the average final embedding layer described in the first approach [i.e. determine a respective target embeddings vector corresponding to a weighted average of the attention embeddings vectors;].”
2. wherein the predicted next user action is determined based on target item embeddings vectors.
(Kulkarni, ¶0112)
“For example, the user activity sequence system 104 may, for each activity event in a sequence of activity events, determine an activity event vector of length 16 [i.e. wherein the predicted next user action is determined based on target item embeddings vectors].”
One of ordinary skill, at the time the invention was filed, would have been motivated to modify Kulkarni with Kang. The motivation is the same as claim 1.
Claims 5, 7-9, 13, 15, 18, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over US Pre-Grant Patent 2022/0107852 (Kulkarni et al; Kulkarni) in view of “Self-Attentive Sequential Recommendation,” arXiv, Kang et al; Kang, 2018, further in view of “Personalized Graph Neural Networks with Attention Mechanism for Session-Aware Recommendation,” arXiv, Zhang et al; Zhang, 2021.
Regarding claim 5:
Kulkarni and Kang teach the method of claim 1.
Neither Kulkarni nor Kang teaches:
1. applying the first embeddings vector index as keys and values, and applying the second embeddings vector index as queries,
2. wherein the attention embeddings vectors are generated based on the first embeddings vector index and the second embeddings vector index.
Zhang teaches:
1. applying the first embeddings vector index as keys and values, and applying the second embeddings vector index as queries,
(Zhang, pg. 5, Sect. 3.5.1, col. 1, ¶5)
“The scaled dot-product attention is formally defined as: where Q, K, V represent the queries, keys, and values respectively [i.e. as keys and values,... as queries].”
(Zhang, pg. 5, Sect. 3.5.1, col. 2, ¶1)
“For current session, we simply denote the embedding matrix as H𝑢 h1h2 h𝑚 . In our context, we use current session embedding [i.e. applying the first embeddings vector index] to query historical session embedding [i.e. and applying the second embeddings vector index], where the queries Q are determined by H𝑢, the keys and values are determined by F𝑢.
2. wherein the attention embeddings vectors are generated based on the first embeddings vector index and the second embeddings vector index.
(Zhang, pg. 5, Sect. 3.5.1, col. 2, ¶1. Eq. 14)
“In specical, we project F𝑢 and H𝑢 to the same latent space through nonlinear transformation [i.e. wherein the attention embeddings vectors are generated based on the first embeddings vector index and the second embeddings vector index].”
PNG
media_image5.png
77
354
media_image5.png
Greyscale
One of ordinary skill, at the time the invention was filed, would have been motivated to modify Kulkarni and Kang with Zhang. The motivation is to modify Kulkarni’s prediction system, as implemented with Kang’s attention mechanism, to use Zhang’s historical-session keys and values and current-session queries. This modification would weight historical information according to its relevance to current behavior, allowing the prediction to reflect both established preferences and present intent. A-PGNN expressly identifies inadequate treatment of historical-session effects as a deficiency and uses transformer attention to “model the explicit effect of historical session on each item of the current session (Zhang, §1, p. 2).”
Regarding claim 7:
Kulkarni and Kang teach the method of claim 1.
Kulkarni teaches:
1. determining a plurality of predicted user actions based on the attention embeddings vectors;
(Kulkarni, ¶0028)
“In particular, an activity event can include any user action taken on a digital content item associated with a user's account and/or a group of user accounts.”
(Kulkarni, ¶0040)
“As mentioned above, the user activity sequence system 104 can predict activity events and, in response, perform an action [i.e. determining a plurality of predicted user actions based on the attention embeddings vectors;].”
2. generating a respective candidate embeddings vector based on the plurality of predicted user actions;
(Kulkarni, ¶0076)
“At act 306, the user activity sequence system 104 can generate a feature vector embedding for each feature extracted at act 304 [i.e. generating a respective candidate embeddings vector based on the plurality of predicted user actions;].”
Neither Kulkarni nor Kang teaches:
1. and determining a respective similarity between the candidate embeddings vectors and the target embeddings vectors;
2. wherein the predicted next user action is determined based on the similarities between the target embeddings vectors and the candidate embeddings vectors.
Zhang teaches:
1. and determining a respective similarity between the candidate embeddings vectors and the target embeddings vectors;
(Zhang, pg. 6, col. 2, Sect. 4.2, ¶4)
“Item-KNN[17] computes an item-to-item cosine similarity based on the co-occurrence of items within sessions [i.e. and determining a respective similarity between the candidate embeddings vectors and the target embeddings vectors;].”
2. wherein the predicted next user action is determined based on the similarities between the target embeddings vectors and the candidate embeddings vectors.
(Zhang, pg. 6, Sect. 3.6, col. 1, ¶2-¶3)
“After obtaining the unified representation of user 𝑢, we compute the recommendation score z𝑖 for each item 𝑣𝑖 ∈ 𝑉. The score function is defined as:
PNG
media_image6.png
19
369
media_image6.png
Greyscale
Where z𝑢 and e𝑣𝑖 denote the user’s unified representation and item 𝑣𝑖’s embedding, respectively. Then we apply a softmax function to get the output vector:
PNG
media_image7.png
24
339
media_image7.png
Greyscale
where ˆz 𝑉 denotes the recommendation scores overall candidate items 𝑉 and ˆy denotes the probabilities that items will be interacted by user 𝑢 in the next time of the current session [i.e. wherein the predicted next user action is determined based on the similarities between the target embeddings vectors and the candidate embeddings vectors].”
One of ordinary skill, at the time the invention was filed, would have been motivated to modify Kulkarni and Kang with Zhang. The motivation is the same as claim 5.
Regarding claim 8:
Kulkarni and Kang teach the method of claim 1.
Zhang teaches:
1. determining a respective similarity score between the target embeddings vectors and the candidate embeddings vectors based on applying a SoftMax loss function;
(Zhang, pg. 6, Sect. 3.6, col. 1, ¶2-¶3, Eq. 22-23)
“After obtaining the unified representation of user 𝑢, we compute the recommendation score z𝑖 for each item 𝑣𝑖 ∈ 𝑉. The score function is defined as [i.e. determining a respective similarity score between the target embeddings vectors]:
PNG
media_image6.png
19
369
media_image6.png
Greyscale
Where z𝑢 and e𝑣𝑖 denote the user’s unified representation and item 𝑣𝑖’s embedding, respectively. Then we apply a softmax function to get the output vector:
PNG
media_image7.png
24
339
media_image7.png
Greyscale
where ˆz 𝑉 denotes the recommendation scores overall candidate items 𝑉 and ˆy denotes the probabilities that items will be interacted by user 𝑢 in the next time of the current session [i.e. and the candidate embeddings vectors based on applying a SoftMax loss function;]
2. and applying the respective similarity score to the respective target embeddings vector.
(Zhang, pg. 6, Sect. 3.6, col. 1, ¶4, Eq. 24)
“For any user behavior graph, the loss function is defined as the cross-entropy of the prediction and the ground truth. It can be written as follows,
PNG
media_image8.png
39
337
media_image8.png
Greyscale
Where y denotes the one-hot encoding vector of the ground truth item [i.e. and applying the respective similarity score to the respective target embeddings vector].”
One of ordinary skill, at the time the invention was filed, would have been motivated to modify Kulkarni and Kang with Zhang. The motivation is the same as claim 5.
Regarding claim 9:
Kulkarni and Kang teach the method of claim 1.
Zhang teaches:
1. wherein the predicted next user action is determined based on the respective similarity score between the target embeddings vectors and the candidate embeddings vectors.
(Zhang, pg. 3, Sect. 3.2, ¶2)
“Using the representations, we output probability ˆy for all candidate items, where the element 𝑦𝑖 ˆy is the recommendation score of the corresponding item 𝑣𝑖 𝑉 (Section 3.6). The items with top-𝑘 values will be the candidate items for recommendation [i.e. wherein the predicted next user action is determined based on the respective similarity score between the target embeddings vectors and the candidate embeddings vectors].”
One of ordinary skill, at the time the invention was filed, would have been motivated to modify Kulkarni and Kang with Zhang. The motivation is the same as claim 5.
Regarding claim 13 and analogous claim 18:
Kulkarni and Kang teach the system of claim 11.
Kang teaches:
1. apply a causal mask to the respective embeddings vector for each of the user actions;
(Kang, pg. 3, Sect. III.B, col.2, ¶3)
“However, the t-th output of the self-attention layer (St) contains embeddings of subsequent items, which makes the model ill-posed. Hence, we modify the attention by forbidding all links between Qi and Kj (j > i) [i.e. applying a causal mask to the respective embeddings vector for each of the user actions;].”
2. and concatenate the first embeddings vector index and the second embeddings vector index to determine the respective pairwise similarities among the embeddings vectors;
(Kang, pg. 3, Sect. III.A, col. 1, ¶3)
“Hence we inject a learnable position embedding P∈Rn×d into the input embedding: [i.e. and concatenating the respective embeddings vectors to determine the respective pairwise similarities among the embeddings vectors].
PNG
media_image4.png
81
346
media_image4.png
Greyscale
Examiner notes that Eq. 1 stacks individual embedding vectors into the input matrix, which fulfills the concatenation requirement. See attached NPL: Concatenation — The Science of Machine Learning & AI.
Zhang teaches:
1. apply the first embeddings vector index as keys and values; apply the second embeddings vector index as queries;
(Zhang, pg. 5, Sect. 3.5.1, col. 1, ¶5)
“The scaled dot-product attention is formally defined as: where Q, K, V represent the queries, keys, and values respectively [i.e. as keys and values,... as queries].”
(Zhang, pg. 5, Sect. 3.5.1, col. 2, ¶1)
“For current session, we simply denote the embedding matrix as H𝑢 h1h2 h𝑚 . In our context, we use current session embedding [i.e. applying the first embeddings vector index] to query historical session embedding [i.e. and applying the second embeddings vector index], where the queries Q are determined by H𝑢, the keys and values are determined by F𝑢.”
2. wherein the attention embeddings vectors are generated based on the first embeddings vector index and the second embeddings vector index.
(Zhang, pg. 5, Sect. 3.5.1, col. 2, ¶1. Eq. 14)
“In specical, we project F𝑢 and H𝑢 to the same latent space through nonlinear transformation [i.e. wherein the attention embeddings vectors are generated based on the first embeddings vector index and the second embeddings vector index].”
PNG
media_image5.png
77
354
media_image5.png
Greyscale
One of ordinary skill, at the time the invention was filed, would have been motivated to modify Kulkarni and Kang with Zhang. The motivation is the same as claim 5.
Regarding claim 15 and analogous claim 20:
Kulkarni and Kang teach the system of claim 11.
Kulkarni teaches:
1. determining a plurality of predicted user actions based on the attention embeddings vectors;
(Kulkarni, ¶0028)
“In particular, an activity event can include any user action taken on a digital content item associated with a user's account and/or a group of user accounts.”
(Kulkarni, ¶0040)
“As mentioned above, the user activity sequence system 104 can predict activity events and, in response, perform an action [i.e. determining a plurality of predicted user actions based on the attention embeddings vectors;].”
2. generating a respective candidate embeddings vector based on the plurality of predicted user actions;
(Kulkarni, ¶0076)
“At act 306, the user activity sequence system 104 can generate a feature vector embedding for each feature extracted at act 304 [i.e. generating a respective candidate embeddings vector based on the plurality of predicted user actions;].”
Zhang teaches:
1. determine a respective similarity score between the target embeddings vectors and the candidate embeddings vectors based on applying a SoftMax loss function;
(Zhang, pg. 6, Sect. 3.6, col. 1, ¶2-¶3)
“After obtaining the unified representation of user 𝑢, we compute the recommendation score z𝑖 for each item 𝑣𝑖 ∈ 𝑉. The score function is defined as:
PNG
media_image6.png
19
369
media_image6.png
Greyscale
Where z𝑢 and e𝑣𝑖 denote the user’s unified representation and item 𝑣𝑖’s embedding, respectively. Then we apply a softmax function to get the output vector:
PNG
media_image7.png
24
339
media_image7.png
Greyscale
where ˆz 𝑉 denotes the recommendation scores overall candidate items 𝑉 and ˆy denotes the probabilities that items will be interacted by user 𝑢 in the next time of the current session.”
2. and apply the respective similarity score to the respective target embeddings vector;
(Zhang, pg. 6, Sect. 3.6, col. 1, ¶4, Eq. 24)
“For any user behavior graph, the loss function is defined as the cross-entropy of the prediction and the ground truth. It can be written as follows,
PNG
media_image8.png
39
337
media_image8.png
Greyscale
Where y denotes the one-hot encoding vector of the ground truth item [i.e. and apply the respective similarity score to the respective target embeddings vector;].”
3. wherein the predicted next user action is determined based on the respective similarity score between the target embeddings vectors and the candidate embeddings vectors.
(Zhang, pg. 3, Sect. 3.2, ¶2)
“Using the representations, we output probability ˆy for all candidate items, where the element 𝑦𝑖 ˆy is the recommendation score of the corresponding item 𝑣𝑖 𝑉 (Section 3.6). The items with top-𝑘 values will be the candidate items for recommendation [i.e. wherein the predicted next user action is determined based on the respective similarity score between the target embeddings vectors and the candidate embeddings vectors].”
One of ordinary skill, at the time the invention was filed, would have been motivated to modify Kulkarni and Kang with Zhang. The motivation is the same as claim 5.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
-US Pre-Grant Patent 2022/0237682 (Zhao et al; Zhao)
-US Pre-Grant Patent 2020/0401661 (Kota et al; Kota)
Any inquiry concerning this communication or earlier communications from the examiner should be directed to PAUL JUSTIN BREENE whose telephone number is (571)272-6320. 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, Michael J Huntley can be reached on 303-297-4307. 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.
/P.J.B./ Examiner, Art Unit 2129
/MICHAEL J HUNTLEY/Supervisory Patent Examiner, Art Unit 2129
