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 .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 02/23/2026 has been entered.
Response to Arguments
Applicant’s argument filed 02/23/2026 have been fully considered but they are not persuasive.
Applicant’s Argument: On page 9-11 of Applicant’s response to rejections under 35 U.S.C. 101, applicant states that the claims are directed to a specific technological improvement in the training of sequential recommendation models and the claim as a whole is not directed to an abstract idea.
The claimed limitation “randomly selecting a subset of items from the plurality of items, wherein a quantity of items in the subset of items is based on a variable, the variable gradually changing throughout a training stage” is regarded as the technical improvement similar to Ex Parte Desjardins. Applicant argues that the claimed dynamic subset selection that prioritizes informative negatives constitutes an improvement to the machine learning model’s training process.
Examiner’s Response: Applicant’s argument is not persuasive. During examination, the examiner should analyze the "improvements" consideration by evaluating the specification and the claims to ensure that a technical explanation of the asserted improvement is present in the specification, and that the claim reflects the asserted improvement (see MPEP §2106.05(a)). The MPEP (§2106.05(a)(II)) also warns, “it is important to keep in mind that an improvement in the abstract idea itself (e.g. a recited fundamental economic concept) is not an improvement in technology.” Here, the alleged improvement in the form of “randomly selecting a subset of items from the plurality of items, wherein a quantity of items in the subset of items is based on a variable, the variable gradually changing throughout a training stage of the sequential recommendation model” is an improvement to the abstract idea of a mental process that can be performed in the human mind.
An important consideration in determining whether a claim improves technology is the extent to which the claim covers a particular solution to a problem or a particular way to achieve a desired outcome, as opposed to merely claiming the idea of a solution or outcome (see MPEP 2106.05(a)). The amended claims do not provide sufficient details to describe any technological improvement. If the specifications explicitly set forth an improvement but in a conclusory manner (see MPEP 2106.04(d)(1): a bare assertion of an improvement without the detail necessary to be apparent to a person of ordinary skill in the art), the examiner should not determine the claim improves technology.
Applicant’s Argument: On page 11-13 of Applicant’s response to rejections under 35 U.S.C. 103, applicant states the recited references failed to teach randomly selecting a subset of items based on a variable that gradually changes during training.
Examiner’s Response: Applicant’s argument is not persuasive. Applicant’s arguments with respect to claim 1 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. The 103 rejections have been updated to reflect the newly added claim limitations.
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 1-7 and 9-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding Claim 1:
Subject Matter Eligibility Analysis Step 1:
Claim 1 recites “A method for training a sequential recommendation model via contrastive learning, comprising” and is thus a process, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
“encoding, s up to a first time instance into a first user interest representation” (a mathematical calculation, par. 20-21 of Specification)
“generating, s” (a mental process that can be performed in the human mind with the aid of pen and paper, i.e. judgement)
“selecting a positive sample corresponding to a next interacted item at a next time instance following the first time instance from the training dataset of user behavior sequences” (a mental process that can be performed in the human mind, i.e. judgement)
“randomly selecting a subset of items from the plurality of items, wherein a quantity of items in the subset of items is based on a variable, the variable gradually changing throughout a training stage of the sequential recommendation model” (a mental process that can be performed in the human mind, i.e. judgement)
“sampling a negative sample from the subset of items according to the first plurality of probabilities, with a highest sampling probability for a particular negative sample being associated with the highest probability of the first plurality of probabilities when the particular negative sample is not the positive sample” (a mental process that can be performed in the human mind, i.e. judgement)
“computing a contrastive loss in response to the inputting” (a mathematical calculation, par. 22 and 24 in Specification)
Claim 1 therefore recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
"receiving, via a communication interface, a training dataset of user behavior sequences” (This step is directed to data gathering, which is understood to be insignificant extra solution activity - see MPEP 2106.05(g))
“encoding, by an encoder of the sequential recommendation model, ” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“generating, by a decoder based on the first user interest representation, ” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
"inputting the sampled negative sample, and the selected positive sample to the sequential recommendation model” (This step is directed to data gathering, which is understood to be insignificant extra solution activity - see MPEP 2106.05(g))
“updating the sequential recommendation model based on the contrastive loss” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
The additional elements as disclosed above alone or in combination do not integrate the judicial exception into practical application as they are mere insignificant extra solution activity in combination of generic computer functions being implemented with generic computer elements in a high level of generality to perform the disclosed abstract idea above. Therefore, Claim 1 is directed to the abstract idea.
Subject Matter Eligibility Analysis Step 2B:
"receiving, via a communication interface, a training dataset of user behavior sequences” (This step is directed to transmitting or receiving information, which is understood to be insignificant extra solution activity and well understood, routine and conventional activity of transmitting and receiving data as identified by the court - see MPEP 2106.05(d))
“encoding, by an encoder of the sequential recommendation model, ” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“generating, by a decoder based on the first user interest representation, ” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
"inputting the sampled negative sample, and the selected positive sample to the sequential recommendation model” (This step is directed to transmitting or receiving information, which is understood to be insignificant extra solution activity and well understood, routine and conventional activity of transmitting and receiving data as identified by the court - see MPEP 2106.05(d))
“updating the sequential recommendation model based on the contrastive loss” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
The additional elements as disclosed above alone or in combination do not recite significantly more than the abstract idea itself as they are mere insignificant extra solution activity in combination of generic computer functions being implemented with generic computer elements in a high level of generality to perform the disclosed abstract idea above. Therefore, Claim 1 is subject-matter ineligible.
Regarding Claim 12:
The claim recites a system (“A system for sequential recommendation, the system comprising”) that performs the method as described in claim 1. Therefore, claim 12 is rejected for the same reasons as disclosed for claim 1. The limitations for additional elements of claim 12 are analyzed below.
Subject Matter Eligibility Analysis Step 2A Prong 1:
Please see Step 2A Prong 1 analysis of claim 1
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“a memory that stores a sequential recommendation model” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“a communication interface that receives a plurality of user behavior sequences” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“one or more hardware processors that” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Regarding Claim 18:
The claim recites an article of manufacture that performs the method as described in claim 1. Therefore, claim 18 is rejected for the same reasons as disclosed for claim 1. The limitations for additional elements of claim 18 are analyzed below.
Subject Matter Eligibility Analysis Step 2A Prong 1:
Please see Step 2A Prong 1 analysis of claim 1
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“A processor-readable non-transitory storage medium storing a plurality of processor- executable instructions for a sequential recommendation model, the instructions being executed by a processor to perform operations comprising” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Regarding Claims 2, 13, and 19:
Subject Matter Eligibility Analysis Step 2A Prong 1:
“computing a distance in a feature space between the user interest representation and representations of the plurality of items” (a mathematical calculation)
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B: None
Regarding Claims 3, 14, and 20:
Subject Matter Eligibility Analysis Step 2A Prong 1:
“computing a first distance in a feature space between a representation of the sampled negative sample and the first user interest representation” (a mathematical calculation)
“computing a second distance in feature space between a representation of the selected positive sample and the first user interest representation” (a mathematical calculation)
“computing the contrastive loss based at least in part on the first distance and the second distance” (a mathematical calculation, par. 22 and 24 in Specification)
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B: None
Regarding Claims 4 and 15:
Subject Matter Eligibility Analysis Step 2A Prong 1: None
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“wherein a first particular item of the plurality of items associated with a higher probability of the first plurality of probabilities than a second particular item of the plurality of items, has a higher probability of being sampled” (merely specifies a particular technological environment in which the abstract idea is to take place, ie. a field of use, and thus does not integrate the abstract idea into a practical application nor cannot provide significantly more than the abstract idea itself - see MPEP 2106.05(h))
Regarding Claims 5 and 16:
Subject Matter Eligibility Analysis Step 2A Prong 1:
“wherein the sampling the negative sample is constrained from sampling the next interacted item from the first sequence of user behaviors” (a mental process that can be performed in the human mind, i.e. judgement)
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B: None
Regarding Claims 6 and 17:
Subject Matter Eligibility Analysis Step 2A Prong 1:
“wherein the updating the sequential recommendation model comprises updating the encoder based on the contrastive loss” (a mathematical calculation; updating based on calculation of the contrastive loss)
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B: None
Regarding Claim 7:
Subject Matter Eligibility Analysis Step 2A Prong 1:
“scaling the first plurality of probabilities based on a scaling parameter” (a mathematical calculation, par. 53 of Specification)
“sampling the negative sample according to scaled probabilities” (a mental process that can be performed in the human mind, i.e. judgement)
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B: None
Regarding Claim 9:
Subject Matter Eligibility Analysis Step 2A Prong 1:
“sampling a plurality of negative samples per one next item prediction at one training time step” (a mental process that can be performed in the human mind, i.e. judgement)
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B: None
Regarding Claim 10:
Subject Matter Eligibility Analysis Step 2A Prong 1: None
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“after updating the sequential recommendation model based on the contrastive loss: re-using the first sequence of user behaviors for training the updated sequential recommendation model at a next training timestep” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Regarding Claim 11:
Subject Matter Eligibility Analysis Step 2A Prong 1:
“including the next interacted item into the first sequence of user behaviors resulting in a second sequence of user behaviors” (a mental process, i.e. judgement)
“encoding the second sequence of user behaviors into a second user interest representation” (a mathematical calculation, par. 20-21 of Specification)
“generating a second plurality of probabilities corresponding to the plurality of items being sequentially recommended as a next item following the second sequence of user behaviors” (a mental process, i.e. judgement)
“sampling another negative sample from the plurality of items according to the second plurality of probabilities” (a mental process, i.e. judgement)
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“using the other negative sample for contrastive learning with the updated sequential recommendation model” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
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 1-3, 6, 9-10, 12-14, and 17-20 are rejected under 35 U.S.C. 103 as being unpatentable over Xie, “Adversarial and Contrastive Variational Autoencoder for Sequential Recommendation” in view of Wu, “Rethinking InfoNCE: How Many Negative Samples Do You Need?” and Zhang, “Optimizing Top-N Collaborative Filtering Via Dynamic Negative Item Sampling”. Xie is cited in PTO-892 dated 06/18/2025 and Zhang is cited in PTO-892 dated 11/25/2025.
Regarding claim 1, Xie teaches:
“A method for training a sequential recommendation model via contrastive learning, comprising” (abstract, A sequential recommendation system is trained by employing contrastive loss.)
“receiving, via a communication interface, a training dataset of user behavior sequences” ([pg. 3, section 3.1, par.1; pg. 6, section 4.1, par. 1-2], The MovieLens Latest and Yelp dataset are obtained to evaluate the model. It is inherent that the datasets are obtained through some communication interface on the computer. The dataset is preprocessed to generate the interaction sequence for each user and items (user behavior), which may include ratings with detailed timestamps (sequences).)
“encoding, by an encoder of the sequential recommendation model, a first sequence of user behaviors up to a first time instance into a first user interest representation” ([pg. 3, section 3, par. 1; pg. 3, section 3.1, par. 1; pg. 3, section 3.2, par. 3; pg.5, section 3.4, par. 2-4; pg. 6, section 4.1, par. 1-2], Sequential variational autoencoder models the influence of past data points on future ones in a time series. The items from 1 to t at timestep t are encoded by the encoder. The dataset may include Yelp reviews by users that consist of ratings and timestamps. Therefore, each interaction sequence between user and items includes a timestamp. The model predicts the next item that the user may be interested in given the user’s historical interaction sequence.)
“generating, by a decoder based on the first user interest representation, a first plurality of probabilities corresponding to a plurality of items being sequentially recommended as a next item following the first sequence of user behaviors” ([pg. 3-4, section 3.2, par. 3-5], An approximate posterior distribution (first plurality of probabilities) is generated based on the latent variable according to the previous items. The target sequence can be generated by the decoder by sampling the target user item recommendation from the distribution.)
“selecting a positive sample corresponding to a next interacted item at a next time instance following the first time instance from the training dataset of user behavior sequences” ([pg. 3, section 3.2, par. 3; pg. 6, section 3.5.1, par. 2], A positive pair of a different user and latent representation is selected based on the posterior distribution (plurality of probabilities). SVAE is recursive and generates a target user item at every timestep t. It is implied that the model is trained for multiple iterations and the later iterations will include the next interacted item at a next time instance following the first time instance.)
“” ([pg. 7, section 4.3, par. 2], The experiment set several fixed sequence lengths M for each dataset to gather different sequences in one batch. If the sequence is larger than M, only the last M items are kept. M is a variable that defines the number of items to keep for each sequence in each dataset. The training of the sequential recommendation model is disclosed.)
“sampling a subset of items according to the first plurality of probabilities, ” ([pg. 3, section 3.2, par. 4-5; pg. 6, section 3.5.1, par. 1-3], The target sequence can be generated by the decoder and the output
x
^
u
is sampled from multinomial distribution.)
“inputting the sampled ” ([pg. 6, section 3.5.1, par. 1-2], The contrastive discriminator receives the positive and negative samples from different users to compute a contrastive loss.)
“computing a contrastive loss in response to the inputting” ([pg. 6, section 3.5.1, par. 3-4], The contrastive discriminator receives the positive and negative samples from different users to compute a contrastive loss.)
“updating the sequential recommendation model based on the contrastive loss” ([pg. 6, section 3.5.2, par. 1], The SVAE is optimized with contrastive learning by minimizing the contrastive loss term of the objective function. The parameters of the encoder are updated.)
Xie does not explicitly disclose an implementation of “randomly selecting a subset of items from the plurality of items, wherein a quantity of items in the subset of items is based on a variable, the variable gradually changing throughout a training stage of the sequential recommendation model” and “sampling a negative sample from the subset of items according to the first plurality of probabilities, with a highest sampling probability for a particular negative sample being associated with the highest probability of the first plurality of probabilities when the particular negative sample is not the positive sample”. However, Wu discloses in the same field of endeavor:
“randomly selecting a subset of items from the plurality of items, wherein a quantity of items in the subset of items is based on a variable, the variable gradually changing throughout a training stage of the sequential recommendation model” ([pg. 1-2, Section 1, par. 3; pg. 3-4, Section 3, par. 3; pg. 7-8, Section 5, par. 1-2], The total number of training samples is separated into different groups based on the model prediction and label to determine training effectiveness. The training effectiveness is a function of the negative sampling ratio. The negative sampling ratio is adjusted during training to determine a set of negative samples associated with each positive sample.)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “randomly selecting a subset of items from the plurality of items, wherein a quantity of items in the subset of items is based on a variable, the variable gradually changing throughout a training stage of the sequential recommendation model” from Wu into the teaching of Xie. Doing so can improve the model training by dynamically adjusting the negative sampling ratio to maximize the training effectiveness function (Wu, abstract).
Xie in view of Wu does not explicitly disclose an implementation of “sampling a negative sample from the subset of items according to the first plurality of probabilities, with a highest sampling probability for a particular negative sample being associated with the highest probability of the first plurality of probabilities when the particular negative sample is not the positive sample”. However, Zhang discloses in the same field of endeavor:
“sampling a negative sample from the subset of items according to the first plurality of probabilities, with a highest sampling probability for a particular negative sample being associated with the highest probability of the first plurality of probabilities when the particular negative sample is not the positive sample” ([pg. 2-3, section 3.1, par. 1-2; pg. 2, Figure 1], A negative item sampling strategy is proposed and shown in Figure 1. The item pairs whose negative item has a higher ranking should be sampled with higher probability. The negative sample is drawn from the unobserved item set and the negative sample is not the positive sample. The unobserved item set is a subset of all the items in the dataset.)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “sampling a negative sample from the subset of items according to the first plurality of probabilities, with a highest sampling probability for a particular negative sample being associated with the highest probability of the first plurality of probabilities when the particular negative sample is not the positive sample” from Zhang into the teaching of Xie in view of Wu. Wu teaches an adaptive negative sampling ratio to determine a random set of negative samples for each positive sample and Zhang teaches negative items sampling based on a ranked list. Doing so can improve the model performance by dynamically choosing negative training samples from a ranked list to update the model (Zhang, abstract).
Regarding claim 12:
Claim 12 recites a system that performs the same process as described in Claim 1. Therefore claim 12 is rejected under the same reasons mention for claim 1. The additional elements of claim 12 is addressed below by Xie:
“a memory that stores a sequential recommendation model” ([pg. 7, section 4.3, par. 1-3], The source code is provided for the sequential recommendation model. It is inherent that the model is stored on a memory component on a computer to perform the methods described.)
“a communication interface that receives a plurality of user behavior sequences” ([pg. 6, section 4.1, par. 1-2], The MovieLens Latest and Yelp dataset are obtained to evaluate the model. It is inherent that the datasets are obtained by an interface on a computer.)
“one or more hardware processors that” ([pg. 7, section 4.3, par. 1-3], It is implied that the source code is performed on a computer consisting of processors to be able to run the code of the model.)
Regarding claim 18:
Claim 18 recites an article of manufacture that performs the same process as described in Claim 1. Therefore claim 18 is rejected under the same reasons mention for claim 1. The additional elements of claim 18 is addressed below by Xie:
“A processor-readable non-transitory storage medium storing a plurality of processor-executable instructions for a sequential recommendation model, the instructions being executed by a processor to perform operations comprising:” ([pg. 7, section 4.3, par. 1-3], The source code is provided for the sequential recommendation model. It is inherent that the model is stored on a memory component on a computer to perform the methods described.)
Regarding claims 2, 13, and 19 Xie teaches:
“computing a distance in a feature space between the user interest representation and representations of the plurality of items” ([pg. 4, section 3.3.2, par. 1], The framework includes adversarial learning in the sequential recommendation model and it computes a KL divergence. The KL divergence computes the distance between the data distributions of all items and the distribution of all items in latent representation according to the last term in equation 4.)
Regarding claims 3, 14, and 20 Xie teaches:
“computing a first distance in a feature space between a representation of the sampled negative sample and the first user interest representation” ([pg. 6, section 3.5.1, par. 1-4, Equation 12], Contrastive loss calculates the Euclidean distance or cosine similarity between vector pairs. The second term of equation 12 computes the similarity between the negative sample pair with the posterior distribution.)
“computing a second distance in feature space between a representation of the selected positive sample and the first user interest representation” ([pg. 6, section 3.5.1, par. 1-4, Equation 12], Contrastive loss calculates the Euclidean distance or cosine similarity between vector pairs. The first term of equation 12 computes the similarity between the positive sample pair with the posterior distribution.)
“computing the contrastive loss based at least in part on the first distance and the second distance” ([pg. 6, section 3.5.1, par. 1-4, Equation 12], Contrastive loss calculates the Euclidean distance or cosine similarity between vector pairs. Equation 12 computes the contrastive loss to distinguish between positive and negative pairs.)
Regarding claims 6, and 17 Xie teaches:
“wherein the updating the sequential recommendation model comprises updating the encoder based on the contrastive loss” ([pg. 6, section 3.5.2, par. 1; pg. 6, section 3.6, par. 1], Contrastive loss is used to optimize the encoder. The objective function consists of the contrastive loss term and may be used to update both the encoder and discriminator.)
Regarding claim 9, Xie teaches:
“sampling a plurality of negative samples per one next item prediction at one training time step” ([pg. 3, section 3.2, par. 3-5; pg. 6, section 3.5.1, par. 2-4], SVAE generates the target user item at every timestep t. SVAE predicts the next item at timestep t. Negative samples are selected from the latent distribution of different users. In Equation 12, Tu represents the total number of items in the user sequence and multiple sample pair is selected to perform the contrastive learning.)
Regarding claim 10, Xie teaches:
“re-using the first sequence of user behaviors for training the updated sequential recommendation model at a next training timestep” ([pg. 3, section 3.2, par. 3-5; pg. 6, Algorithm 1], SVAE generates the target user item at every timestep t. SVAE predicts the next item at timestep t. For each iteration of running the model to predict the next item recommendation, the model uses the previous items in making a prediction.)
Claims 4 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over Xie, “Adversarial and Contrastive Variational Autoencoder for Sequential Recommendation” in view of Wu, “Rethinking InfoNCE: How Many Negative Samples Do You Need?”, Zhang, “Optimizing Top-N Collaborative Filtering Via Dynamic Negative Item Sampling” and Hao, “Adversarial Feature Translation for Multi-domain Recommendation”. Hao is cited in PTO-892 dated 06/18/2025.
Regarding claims 4, and 15 Xie in view of Wu and Zhang teaches:
“wherein a first particular item of the plurality of items associated with a higher probability of the first plurality of probabilities than a second particular item of the plurality of items, ” ([Xie, pg. 3, section 3.3, par. 3-5], The recommended item for a user is generated by sampling from a multinomial distribution. A multinomial distribution models different probabilities that are assigned to each category. It is implied that the distribution of items has different probabilities of being selected although it is not explicitly recited in the reference.)
Xie in view of Wu and Zhang does not explicitly disclose an implementation of “wherein a first particular item of the plurality of items associated with a higher probability of the first plurality of probabilities than a second particular item of the plurality of items, has a higher probability of being sampled”. However, Hao discloses in the same field of endeavor:
“wherein a first particular item of the plurality of items associated with a higher probability of the first plurality of probabilities than a second particular item of the plurality of items, has a higher probability of being sampled” ([pg.5, section 3.3.2, par. 4; pg. 5, section 4.1, par. 2], A ranking score is generated for item candidates and items have a higher score if it is most similar to the user domain-specific representation. A top-N candidate can be selected based on the ranking scores.)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “wherein a first particular item of the plurality of items associated with a higher probability of the first plurality of probabilities than a second particular item of the plurality of items, has a higher probability of being sampled” from Hao into the teaching of Xie in view of Wu and Zhang. Doing so can improve recommendations by the system by learning the feature translation between different domains (Hao, abstract).
Claims 5, 11, and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Xie, “Adversarial and Contrastive Variational Autoencoder for Sequential Recommendation” in view of Wu, “Rethinking InfoNCE: How Many Negative Samples Do You Need?”, Zhang, “Optimizing Top-N Collaborative Filtering Via Dynamic Negative Item Sampling” and Ding, “Reinforced Negative Sampling for Recommendation with Exposure Data”. Ding is cited in PTO-892 dated 06/18/2025.
Regarding claims 5, and 16, Xie in view of Wu and Zhang teaches:
“wherein the sampling the negative sample is s” ([Xie, pg. 6, section 3.5.1, par. 1-4], The negative samples selected to compute the contrastive loss is based on historical user interactions. The reference does not explicitly disclose any constraints on sampling of the negative sample.)
Xie in view of Wu and Zhang does not explicitly disclose an implementation of “wherein the sampling the negative sample is constrained from sampling the next interacted item from the first sequence of user behavior”. However, Ding discloses in the same field of endeavor:
“wherein the sampling the negative sample is constrained from sampling the next interacted item from the first sequence of user behavior” ([pg. 2-3, section 2.2, par. 4; pg. 5-6], Real negative samples are generated based on the exposure data (next interacted item) that have not been interacted by the user. The negative samples refer to the non-interacted instances in the exposure or recommended data.)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “wherein the sampling the negative sample is constrained from sampling the next interacted item from the first sequence of user behavior” from Ding into the teaching of Xie in view of Wu and Zhang. Doing so can improve recommendations by the system by optimizing the sampling of high-quality negative samples (Ding, abstract).
Regarding claim 11, Xie in view of Wu teaches:
“including the ” ([Xie, pg. 3, section 3.2, par. 3, pg. 6, Algorithm 1], The model may run for multiple iterations to predict the (t+1)-th item based on the historical user interaction from 1 to t. SVAE is recursive and it is implied that the results of a previous step are used as input for the next step in a continuous cycle.)
“encoding the second sequence of user behaviors into a second user interest representation” ([Xie, pg. 3, section 3.2, par. 3-4; pg.5 , section 3.4, par. 2-4], Sequential variational autoencoder models the influence of past data points on future ones in a time series. The items from 1 to t at timestep t are encoded by the encoder to generate the latent variable hu,t.)
“generating a second plurality of probabilities corresponding to the plurality of items being sequentially recommended as a next item following the second sequence of user behaviors” ([Xie, pg. 3-4, section 3.2, par. 3-5], An approximate posterior distribution (second plurality of probabilities) is generated based on the latent variable according to the previous items. The target sequence can be generated by the decoder by sampling the target user item recommendation from the distribution.)
“sampling another ” ([Xie, pg. 3, section 3.2, par. 4-5], The target sequence can be generated by the decoder and the output
x
^
u
is sampled from multinomial distribution. When the model is executed for additional iterations from the first, the sequential recommender may sample additional items from the new probability distributions.)
“using the other negative sample for contrastive learning with the updated sequential recommendation model” ([Xie, pg. 6, section 3.5.1, par. 2; pg. 6, section 3.5.2, par. 1], The contrastive discriminator receives the positive and negative samples from different users to compute a contrastive loss. Optimization is performed on the contrastive learning to minimize the contrastive loss term of the objective function and the calculation of contrastive loss will be determined for multiple iterations. The model is updated at each iteration.)
Xie in view of Wu does not explicitly disclose an implementation of “sampling another negative sample from the plurality of items according to the second plurality of probabilities” and “including the next interacted item into the first sequence of user behaviors resulting in a second sequence of user behaviors”. However, Zhang discloses in the same field of endeavor:
“sampling another negative sample from the plurality of items according to the second plurality of probabilities” ([pg. 2-3, section 3.1, par. 1-2; pg. 3, Section 3.1.1, par. 1-2; pg. 2, Figure 1], A negative item sampling strategy is proposed and shown in Figure 1. Item in rank 1 and in rank 4 are sampled and the weight are determined for each unobserved item. The sampling probability is a linear function to the relative ranking position.)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “sampling another negative sample from the plurality of items according to the second plurality of probabilities” from Zhang into the teaching of Xie in view of Wu. Doing so can improve the model performance by dynamically choosing negative training samples from a ranked list to update the model (Zhang, abstract).
Xie in view of Wu and Zhang do not explicitly disclose an implementation of “including the next interacted item into the first sequence of user behaviors resulting in a second sequence of user behaviors”. However, Ding discloses in the same field of endeavor:
“including the next ” ([pg. 1, section 1, par. 6; pg. 2, section 2, par. 1-2; Figure 1], The sampler generates hard and real negative instances. The negative instances are sampled based on overlap with non-interacted instances in the exposure data, which contains records of user’s interactions and non-interactions. From Figure 1, the sampled negative instance is provided to the recommender as input for pairwise learning.)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “including the next interacted item into the first sequence of user behaviors resulting in a second sequence of user behaviors” from Ding into the teaching of Xie in view of Wu and Zhang. Doing so can improve recommendations by the system by optimizing the sampling of high-quality negative samples (Ding, abstract).
Claim 7 is rejected under 35 U.S.C. 103 as being unpatentable over Xie, “Adversarial and Contrastive Variational Autoencoder for Sequential Recommendation” in view of Wu, “Rethinking InfoNCE: How Many Negative Samples Do You Need?”, Zhang, “Optimizing Top-N Collaborative Filtering Via Dynamic Negative Item Sampling” and Lee (US20200380583A1). Lee is cited in PTO-892 dated 06/18/2025.
Regarding claim 7, Xie in view of Wu and Zhang teaches:
“sampling the negative sample according to ” ([Zhang, pg. 2-3, section 3.1, par. 1-2; pg. 3, Section 3.1.1, par. 1-2; pg. 2, Figure 1], A negative item sampling strategy is proposed and shown in Figure 1. The sampling probability is a linear function to the relative ranking position.)
Xie in view of Wu and Zhang does not explicitly disclose an implementation of “scaling the first plurality of probabilities based on a scaling parameter” and “sampling the negative sample according to scaled probabilities”. However, Lee discloses in the same field of endeavor:
“scaling the first plurality of probabilities based on a scaling parameter” ([0046-0047, 0049, Figure 7A], A scaling value may be applied to a parameter of the probability distribution to generate a new scaled probability distribution.)
“sampling the ” ([0053-0054], The recommendation system may draw a recommendation from the scaled probability distribution.)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “scaling the first plurality of probabilities based on a scaling parameter” and “sampling the negative sample according to scaled probabilities” from Lee into the teaching of Xie in view of Wu and Zhang. Doing so can improve recommendations by the system by optimizing the sampling of recommendations using a scaling cycle (Lee, abstract).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to GARY MAC whose telephone number is (703)756-1517. The examiner can normally be reached Monday - Friday 8:00 AM - 5:00 PM.
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.
/GARY MAC/Examiner, Art Unit 2127
/ABDULLAH AL KAWSAR/Supervisory Patent Examiner, Art Unit 2127
Prosecution Insights