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 § 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.
Claim(s) 1-2, 4-10, 12-18 and 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over He et al. (“Adversarial Personalized Ranking for Recommendation”, Published: 27 June 2018, pg. 355-364) in view of Kang et al. (“Self-Attentive Sequential Recommendation”, 2018 IEEE).
Regarding claim 1.
He teaches a method for training a sequential recommendation model, the method comprising: see page 363, “Due to the abundance of user feedback such ratings and purchases that can directly reflect a user’s preference, research on item recommendation have mainly focused on mining the feedback data, known as collaborative filtering (CF). Among the various CF methods, matrix factorization (MF) [18], a special type of latent factor models, is a basic yet most effective recommender model.”);
generating, via the sequential recommendation model, a recommendation score relating to a target item for the user (see page 356, “Formally, let u denote a user and i denote an item, then the predictive model of MF is formulated as:
PNG
media_image1.png
28
134
media_image1.png
Greyscale
, where pu ∈ RK and qi ∈ RK denote the embedding vector for user u and item i, respectively, and K is the size of embedding vector also called as embedding size. Θ denotes the model parameters of MF, which is consisted of all user embedding and item embedding vectors, i.e., Θ = {{pu}u∈U, {qi}i∈I}, where U and I denote the set of all users and items, respectively. We use P and Q to denote the embedding matrix P = {pu}u∈U,Q = {qi}i∈I for short.”, i.e. yui== user embedding, qi == recommendation score relating to a target item for the user);
computing a first loss based on the generated recommendation score (see page 356, “Bayesian Personalized Ranking (BPR) maximizes the margin between an observed interaction and its unobserved counterparts. This is fundamentally different from pointwise methods[2, 17] that optimize each model prediction towards a predefined ground truth. Formally, the objective function (to be minimized) of BPR is
PNG
media_image2.png
50
468
media_image2.png
Greyscale
… It is worth noting that the behavior of BPR can be interpreted as a classifier — given a triplet (u,i, j), it determines whether (u,i) should have a higher score than (u,j). ”);
perturbing embeddings of the items based on the respective cascade scores (see page 359, “The solution is simple and straightforward— we first train MF with BPR, and then further optimize it under our APR framework. We term the method as Adversarial Matrix Factorization (AMF). Figure 2 illustrates our AMF method. Since the parameters of MF are embedding vectors for users and items, we apply adversarial perturbations on the embedding vector. Given a (u, i ) pair, the predictive model with perturbations is defined as
PNG
media_image3.png
40
434
media_image3.png
Greyscale
", also see Figure 2: Illustration of our AMF method. The perturbations Δ are enforced on each embedding vector of user and item, also see page 357, section 3.1);
computing a second loss based on the perturbed embeddings of the items (see page 357, “Due to the rationality of the BPR pairwise objective in personalized ranking, we choose it as the building block. To enhance the robustness, we enforce the model to perform well even when the adversarial perturbations (defined in Equation (2)) are presented. To achieve this, we additionally optimize the model to minimize the BPR objective function with the perturbed parameters. Formally, we define the objective function of adversarial personalized ranking as follows:
PNG
media_image4.png
70
426
media_image4.png
Greyscale
”);
and updating parameters of the sequential recommendation model based on the first loss and the second loss via backpropagation (see page 358, “We can see that, similar to BPR, our formulation of APR leads to a general learning framework which is model independent. As long as the underlying model ˆyui(Θ) is differentiable, it can be learned under our APR framework using backpropagation and gradient based optimization algorithms. There are two hyper-parameters — ϵ and λ —to be specified in APR in addition to the ones in BPR. In what follows, we propose a generic solution for APR based on SGD…
PNG
media_image5.png
422
494
media_image5.png
Greyscale
”).
He do not specifically teach receiving, via a data interface, a training dataset comprising a plurality of user interaction sequences for a user; and computing, for at least one of the plurality of user interaction sequences, respective cascade scores for items in the at least one of the plurality of user interaction sequences based on a sequence position and the presence of the items in other sequences of the plurality of user interaction sequences.
Kang teaches receiving, via a data interface, a training dataset comprising a plurality of user interaction sequences for a user (see 198, “Sequential recommendation (or next-item recommendation) differs slightly from this setting, as it only considers the order of actions, and models sequential patterns which are independent of time. Essentially, sequential models try to model the ‘context’ of users’ actions based on their recent activities, rather than considering temporal patterns per se.”, also see page 199, “In the setting of sequential recommendation, we are given a user’s action sequence Su = (Su 1 , Su2 , . . . , Su |Su|), and seek to predict the next item. During the training process, at time step t, the model predicts the next item depending on the previous t items. As shown in Figure 1, it will be convenient to think of the model’s input as (Su 1 , Su 2 , . . . , Su |Su|−1) and its expected output as a ‘shifted’ version of the same sequence: (Su 2 , Su 3 , . . . , Su |Su|).” ); and computing, for at least one of the plurality of user interaction sequences, respective cascade scores for items in the at least one of the plurality of user interaction sequences based on a sequence position and the presence of the items in other sequences of the plurality of user interaction sequences (see page 199, “Positional Embedding: 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…”, also see page 201, “Factorized Item Similarity Models [8]: FISM estimates a preference score toward item i by considering the similarities between i and items the user consumed before…If we use one self-attention layer (excluding the feedforward network), set uniform attention weights (i.e., 1 |Su| ) on items, use unshared item embeddings, and remove the position embedding, SASRec is reduced to FISM. Thus our model can be viewed as an adaptive, hierarchical, sequential item similarity model for next item recommendation.”).
Both He and Kang pertain to the problem of item recommendation systems, thus being analogous. It would have been obvious to one skilled in the art before the effective filing date of the claimed invention to combine He and Kang to teach the above limitations. The motivation for doing so would be “To capture such patterns, two approaches have proliferated: Markov Chains (MCs) and Recurrent Neural Networks (RNNs)… At each time step, SASRec seeks to identify which items are ‘relevant’ from a user’s action history, and use them to predict the next item. Extensive empirical studies show that our method outperforms various state-of-the-art sequential models (including MC/CNN/RNN-based approaches) on both sparse and dense datasets. Moreover, the model is an order of magnitude more efficient than comparable CNN/RNN-based models. Visualizations on attention weights also show how our model adaptively handles datasets with various density, and uncovers meaningful patterns in activity sequences.” (see Kang abstract).
Regarding claim 2.
He and Kang teaches the method of claim 1,
Kang further teaches wherein the respective cascade scores are computed by assigning a higher score to items in earlier positions in their respective user interaction sequences (see page 200, “a high interaction score ri,t means a high relevance, and we can generate recommendations by ranking the scores.”, also see page 201, “Factorized Item Similarity Models [8]: FISM estimates a preference score toward item i by considering the similarities between i and items the user consumed before…If we use one self-attention layer (excluding the feedforward network), set uniform attention weights (i.e., 1 |Su| ) on items, use unshared item embeddings, and remove the position embedding, SASRec is reduced to FISM. Thus our model can be viewed as an adaptive, hierarchical, sequential item similarity model for next item recommendation.”).
The motivation utilized in the combination of claim 1, super, applies equally as well to claim 2.
Regarding claim 4.
He and Kang teaches the method of claim 1,
He further teaches the computing the second loss further comprising:
generating a first user embedding based on embeddings of the items;
generating a second user embedding based on the perturbed embeddings of the items; and
computing the second loss by computing a difference between the first user embedding and the second user embedding (see page 357, “Due to the rationality of the BPR pairwise objective in personalized ranking, we choose it as the building block. To enhance the robustness, we enforce the model to perform well even when the adversarial perturbations (defined in Equation (2)) are presented. To achieve this, we additionally optimize the model to minimize the BPR objective function with the perturbed parameters. Formally, we define the objective function of adversarial personalized ranking as follows:
PNG
media_image4.png
70
426
media_image4.png
Greyscale
”).
Regarding claim 5.
He and Kang teaches the method of claim 1,
He further teaches wherein the perturbing the embeddings of the items includes:
determining a worst case perturbation direction based on a gradient of a function defining the second loss, and perturbing the embeddings of the items based on the determined worst case perturbation direction (see page 358, “We can see that, similar to BPR, our formulation of APR leads to a general learning framework which is model independent. As long as the underlying model ˆyui(Θ) is differentiable, it can be learned under our APR framework using backpropagation and gradient based optimization algorithms. There are two hyper-parameters — ϵ and λ —to be specified in APR in addition to the ones in BPR. In what follows, we propose a generic solution for APR based on SGD.”).
Regarding claim 6.
He and Kang teaches the method of claim 1,
He further teaches further comprising: generating a second recommendation score using at least one of: a perturbed user embedding; a perturbed target item embedding; or a perturbed negative item embedding; and computing a third loss based on the second recommendation score, wherein the updating parameters is further based on the third loss (see page 357, “Due to the rationality of the BPR pairwise objective in personalized ranking, we choose it as the building block. To enhance the robustness, we enforce the model to perform well even when the adversarial perturbations (defined in Equation (2)) are presented. To achieve this, we additionally optimize the model to minimize the BPR objective function with the perturbed parameters. Formally, we define the objective function of adversarial personalized ranking as follows:
PNG
media_image4.png
70
426
media_image4.png
Greyscale
”, i.e. multiple losses are computed and model is evaluated with perturbed parameters).
Regarding claim 7.
He and Kang teaches the method of claim 6,
He further teaches wherein at least one of the perturbed user embedding, perturbed target item, or perturbed negative item embedding is perturbed by determining a worst case perturbation direction based on one or more gradients of a function defining the third loss (see page 357, “Due to the rationality of the BPR pairwise objective in personalized ranking, we choose it as the building block. To enhance the robustness, we enforce the model to perform well even when the adversarial perturbations (defined in Equation (2)) are presented. To achieve this, we additionally optimize the model to minimize the BPR objective function with the perturbed parameters. Formally, we define the objective function of adversarial personalized ranking as follows:
PNG
media_image4.png
70
426
media_image4.png
Greyscale
”).
Regarding claim 8.
He and Kang teaches the method of claim 6,
He further teaches wherein updating the parameters is based on a combined loss objective including a weighted sum of the first loss, the second loss, and the third loss (see page 357, “Due to the rationality of the BPR pairwise objective in personalized ranking, we choose it as the building block. To enhance the robustness, we enforce the model to perform well even when the adversarial perturbations (defined in Equation (2)) are presented. To achieve this, we additionally optimize the model to minimize the BPR objective function with the perturbed parameters. Formally, we define the objective function of adversarial personalized ranking as follows:
PNG
media_image4.png
70
426
media_image4.png
Greyscale
”).
Claims 9-10 and 12-16 recites a system to perform the method recited in claims 1-2 and 4-8. Therefore the rejection of claims 1-2 and 4-8 above applies equally here. Kang also teaches the addition elements of claim 9 not recited in claim 1 comprising a memory that stores the sequential recommendation model and a plurality of processor executable instructions (see Kang, page 197, Figure 1 shows computer etc. and a simplified diagram showing the training process of SASRec. At each time step, the model considers all previous items, and uses attention to ‘focus on’ items relevant to the next action).
Claims 17-18 and 20 recites a non-transitory machine-readable medium to perform the method recited in claims 1-2 and 4. Therefore the rejection of claims 1-2 and 4 above applies equally here. Kang also teaches the addition elements of claim 17 not recited in claim 1 comprising a plurality of machine-executable instructions which, when executed by one or more processors (see Kang, page 197, Figure 1 shows computer etc. and a simplified diagram showing the training process of SASRec. At each time step, the model considers all previous items, and uses attention to ‘focus on’ items relevant to the next action).
Allowable Subject Matter
Claims 3, 11 and 19 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Related prior arts:
Vitaladevuni et al. (US 9443198 B1) teaches conventional approach to detection is to process a sample through a cascade classifier, also referred to herein as a detection cascade or simply as a cascade, that includes a sequence of classifiers that score the sample on how likely it is to contain the detection target. At each stage of the cascade, a decision is made to either discard the sample under consideration or pass it on to the next stage.
Ma et al. (US 20230281448 A1) teaches first recommendation scores of the plurality of recommendation dimensions based on the mapping feature to obtain a fusion feature, and performing recommendation score prediction processing on the to-be-recommended information based on the fusion feature to obtain a second recommendation score of the target object for the to-be-recommended information; and executing a recommendation operation of the to-be-recommended information corresponding to the target object based on the second recommendation score.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to IMAD M KASSIM whose telephone number is (571)272-2958. The examiner can normally be reached 10:30AM-5:30PM, M-F (E.S.T.).
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 at (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.
/IMAD KASSIM/Primary Examiner, Art Unit 2129