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 .
Specification
The disclosure is objected to because of the following informalities: p. Appropriate correction is required.
The disclosure is objected to because of the following informalities: Paragraph [0068] recites "A gating mechanism of a graph neural network (RNN) is applied to node propagation." The acronym "RNN" does not correspond to "graph neural network" (GNN). Appropriate correction is required.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claim 10, 20, and 30 rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 10, 20, and 30 recites the limitation "the item" in line 1 of the claim. There is insufficient antecedent basis for this limitation in the claim. Claim 1, from which claim 10 depends, recites "at least one item," "a sample item," and "a target item," but never a singular "an item." It is therefore unclear which of the three previously recited items the limitation "the item" refers back to, rendering the scope of the claim uncertain.
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, 3, 10-11, 13, 20-21, 23, and 30 is/are rejected under 35 U.S.C. 103 as being unpatentable over Pande et al. (US 20200250734 A1) in view of Wang et al. (US 20210390394 A1).
As to claim 1, Pande teaches an item recommendation method, wherein the method comprises: (see Pande paragraph [0024] "embodiments of the present invention are directed to methods and systems for generating item recommendations using convolutions on weighted graphs.")
obtaining historical interaction data of a target object, wherein the historical interaction data indicates a historical interaction event between the target object and at least one item; (see Pande paragraph [0031] "the recommendation modeling engine 112 performs a process including sampling, weighting, and aggregation of graph-based data, generated from a combination of image data 130, item data 132 (e.g., text descriptions of items), and user selection data 134 (e.g., page or item views, item selections, purchases, etc.).", and see Pande paragraph [0038] "the method 200 can be used to identify items that are in some way related to an initially selected item (as identified by, e.g., prior user activity).", and see Pande paragraph [0048] "The selected item may be an item selected by receiving the item from a retail web server, e.g., in response to a user selecting that item for display from an item collection.")
invoking, based on the historical interaction data, the target recommendation model to output a target item corresponding to the target object. (see Pande paragraph [0036] "The item recommendation subsystem 120 utilizes a model generated from the other subsystems and can receive an identification of an item within an item collection, e.g., from a retail web server 12. The item recommendation subsystem 120 can then utilize the model to identify one or more recommended items in response, which can be provided to the retail web server 12 for presentation to a user", and see Pande paragraph [0048] "Once the weighted loss function is applied, the output of convolution of the sampled weighted graph can be used to generate recommended items (step 210). The recommended items can include one or more item recommendations representing neighbors of a selected item.")
Pande does not explicitly teach "obtaining a pre-trained target recommendation model, wherein the target recommendation model comprises a graph neural network model with one convolutional layer, and the convolutional layer indicates an association relationship between a sample object and a sample item; and"
However, Wang teaches
obtaining a pre-trained target recommendation model, wherein the target recommendation model comprises a graph neural network model with one convolutional layer, and the convolutional layer indicates an association relationship between a sample object and a sample item; and (see Wang paragraph [0035] "a recommendation model to be trained is a model based on GCN. Based on the GCN, convolution can be defined in an irregular graph network structure (e.g., user-recommended content network structure), and then different recommended contents can be determined to be provided to different users through the model.", and see Wang paragraph [0041] "The relation matrix for the users and the recommended contents specifically refers to a matrix used to represent an association relation between the users and the recommended contents", and see Wang paragraph [0051] "Taking one-layer GCN as an example, in a multi-graph GCN, X.sub.Θ is represented as: X.sub.Θ=GCN(X.sup.0;Θ)=relu(L.sub.rX.sup.0L.sub.cΘ)", and see Wang paragraph [0093] "the relation matrix for users and the relation matrix for recommended contents are input into the recommendation model pre-trained by the method for generating a recommendation model of any embodiment of the disclosure")
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the invention to modify the graph-convolutional item recommendation method of Pande to use a pre-trained recommendation model comprising a one-layer graph convolutional network whose convolution indicates the user-item association relation, as taught by Wang, in order to reduce the amount of data and the number of parameters to be learned by the graph convolutional network while preserving fitting accuracy, thereby speeding up training and computation of the recommendation model, as suggested by Wang see paragraph [0066].
As to claim 3, Pande as modified by Wang teaches the method according to claim 1,
wherein the target recommendation model further indicates an association relationship between at least two sample items and/or an association relationship between at least two sample objects. (see Wang paragraph [0037] "The relation matrix for users specifically refers to a matrix used to represent an association relation between different users, which may be generated by a user relation graph.", and see Wang paragraph [0040] "the relation matrix for recommended contents specifically refers to a matrix used to represent an association relation between different recommended contents, and the relation matrix for recommended contents is generated by a recommended content relation graph.", and see Wang paragraph [0042] "By inputting the above-mentioned graph training sample set into the machine learning model for training, the machine learning model can learn the data of the relation matrix for the users and the recommended contents under joint action of different relation matrixes for users and relation matrixes for recommended contents.")
As to claim 10, Pande as modified by Wang teaches the method according to claim 1, wherein the item comprises at least one of text, a link, a product, a service, and concept information. (see Pande paragraph [0024] "such a framework can be used to generate related product recommendations for a retailer to present to a user, e.g., in an online retail environment.")
As to claim 11, this is directed to an apparatus that corresponds to the method of claim 1, See the rejection for claim 1 above, which also applies to claim 11. In addition the claim recited the additional elements
a processor; and (see Pande paragraph [0028] "includes a processor 102")
a memory configured to store instructions that when executed by the processor, cause the apparatus to perform operations of: (see Pande paragraph [0028] "The recommendation modeling computing system 100 includes a processor 102 communicatively connected to a memory 104")
As to claim 13, this is directed to an apparatus that corresponds to the method of claim 3, See the rejection for claim 3 above, which also applies to claim 13.
As to claim 20, this is directed to an apparatus that corresponds to the method of claim 10, See the rejection for claim 10 above, which also applies to claim 20.
As to claim 21, this is directed to a computer program that corresponds to the method of claim 1, See the rejection for claim 1 above, which also applies to claim 21.
As to claim 23, this is directed to a computer program that corresponds to the method of claim 3, See the rejection for claim 3 above, which also applies to claim 23.
As to claim 30, this is directed to a computer program that corresponds to the method of claim 10, See the rejection for claim 10 above, which also applies to claim 30.
Claim(s) 2, 4-9, 12, 14-19, 22, and 24-29 is/are rejected under 35 U.S.C. 103 as being unpatentable over Pande et al. (US 20200250734 A1) in view of Wang et al. (US 20210390394 A1) and Menon et al. (US 20210049442 A1).
As to claim 2, Pande as modified by Wang teaches the method according to claim 1,
Pande does not explicitly teach "wherein the target recommendation model indicates associated item information corresponding to each of n sample objects and associated object information corresponding to each of m sample items, the associated item information indicates at least one item that has interacted with the sample object, the associated object information indicates at least one object that has interacted with the sample item, and n and m are positive integers"
However, Menon teaches
wherein the target recommendation model indicates associated item information corresponding to each of n sample objects and associated object information corresponding to each of m sample items, the associated item information indicates at least one item that has interacted with the sample object, the associated object information indicates at least one object that has interacted with the sample item, and n and m are positive integers. (see Menon paragraph [0027] "A.sup.a,b ... is the adjacency matrix of the graph, the rows and columns of A.sup.a,b being associated with entity instances of types ε.sup.(a) and ε.sup.(b), respectively.", and see Menon paragraph [0031] "The normalized Laplacian for the graph is defined as: L=I−D.sup.−1/2AD.sup.−1/2, where D is the diagonal degree matrix of A", and see Menon paragraph [0054] "user-author graphs, with directional edges such as the number of times a user liked an author's post, replied to an author's post, etc.; and (3) user-channel graphs, with edges capturing interactions such as how many times a user visited a channel in the past month")
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the invention to further modify the recommendation method of Pande as modified by Wang to train the model on labeled user-item pairs whose adjacency and degree structure indicates each user's interacted items and each item's interacting users, as taught by Menon, in order to obtain an accurate recommendation model that identifies and surfaces the items most relevant to each user, as suggested by Menon see paragraph [0053].
As to claim 4, Pande as modified by Wang teaches the method according to claim 1,
wherein the invoking, based on the historical interaction data, the target recommendation model to output a target item corresponding to the target object comprises: (see Pande paragraph [0036] "The item recommendation subsystem 120 utilizes a model generated from the other subsystems and can receive an identification of an item within an item collection, e.g., from a retail web server 12. The item recommendation subsystem 120 can then utilize the model to identify one or more recommended items in response, which can be provided to the retail web server 12 for presentation to a user", and see Pande paragraph [0048] "Once the weighted loss function is applied, the output of convolution of the sampled weighted graph can be used to generate recommended items (step 210). The recommended items can include one or more item recommendations representing neighbors of a selected item.")
inputting the historical interaction data to the target recommendation model, to output the target item corresponding to the target object, (see Pande paragraph [0036] "The item recommendation subsystem 120 utilizes a model generated from the other subsystems and can receive an identification of an item within an item collection, e.g., from a retail web server 12. The item recommendation subsystem 120 can then utilize the model to identify one or more recommended items in response", and see Pande paragraph [0048] "The selected item may be an item selected by receiving the item from a retail web server, e.g., in response to a user selecting that item for display from an item collection. Item recommendations can be returned to the retail web server for display to the user.")
Pande does not explicitly teaches "wherein the target recommendation model is obtained through training based on at least one sample data group, each sample data group comprises sample interaction data and pre-labeled correct recommendation information, and the sample interaction data indicates a sample object and a corresponding sample item."
However, Menon teaches
wherein the target recommendation model is obtained through training based on at least one sample data group, each sample data group comprises sample interaction data and pre-labeled correct recommendation information, and the sample interaction data indicates a sample object and a corresponding sample item. (see Menon paragraph [0041] "The computational models and algorithms used in generating recommendations in accordance herewith, such as the neural-network architecture 300, may be trained based on labeled training data ... where y.sub.i is the label of the i-th pair of first and second items in the dataset", and see Menon paragraph [0054] "In the training data, the label y for any pair … may be a binary indicator of whether the user has "engaged" with the message, where "engaging" may be defined, e.g., as "liking" or replying to a message.")
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the invention to further modify the recommendation method of Pande as modified by Wang so that the recommendation model is obtained through training based on sample data groups of user-item interaction pairs with pre-labeled correct recommendation information, as taught by Menon, in order to obtain a scoring function whose computed relevance scores are consistent with the labeled training data, thereby producing an accurate recommendation model that identifies the items most relevant to each user, as suggested by Menon see paragraph [0041].
As to claim 5, Pande-Wang as modified by Menon teaches the method according to claim 4, further comprising:
before the obtaining a pre-trained target recommendation model, obtaining a training sample set, wherein the training sample set comprises the at least one sample data group; (see Menon paragraph [0050] "The method 600 takes training data, test data, and a specified number of iterations as input, and returns a value of AUC—Rel@k as output.")
for each of the at least one sample data group, inputting the sample interaction data to an initial parameter model to obtain a training result; (see Menon paragraph [0050] "After receiving the input (in act 602) and initializing the neural-network parameters 504, the training dataset, and an iteration counter (in act 604)", and see Menon paragraph [0051] "The trained neural-network architecture 502 is used to score all datapoints in the received training data (act 610)")
obtaining a training loss value through calculation based on the training result and the correct recommendation information by using a preset loss function, wherein the training loss value indicates an error between the training result and the correct recommendation information; (see Menon paragraph [0051] "the neural-network architecture 502 is trained on the current training dataset, in act 608, to minimize an objective function such as, e.g., the regularized form of the negative log likelihood (corresponding to cross entropy) of the predicted score s(U,J)", and see Menon paragraph [0051] "y.sub.i are the labels given to datapoint (U.sub.i,J.sub.i) in the training data, τ is the regularizer, and the second term is a sum of l2-norms of all the weight matrices in the architecture 502.")
adjusting a model parameter in the initial parameter model based on the training loss value corresponding to each of the at least one sample data group; and (see Menon paragraph [0048] "Optimization of the parameters 504 involves executing a learning algorithm 508 to iteratively optimize a surrogate metric 510 for the performance metric 506 based on labeled training data 512 provided as input to the training system 500. The learning algorithm 508 may utilize, e.g., backpropagation of errors")
when a preset convergence condition is met, obtaining the target recommendation model through training. (see Menon paragraph [0051] "When the iteration counter has reached the specified number of iterations, training is deemed complete, and the metric AUC—Rel@k is evaluated on the test data and provided as output (act 616).")
As to claim 6, Pande-Wang as modified by Menon teaches the method according to claim 5,
at least one negative sample data group that is randomly sampled (see Pande paragraph [0047] "P.sub.n is a negative sampling distribution, and Q defines the number of negative samples.")
Pande does not explicitly teach "wherein the at least one sample data group comprises at least one positive sample data group and … , there is an interaction event between the sample object and the sample item in the positive sample data group, and there is no interaction event between the sample object and the sample item in the negative sample data group."
However, Menon teaches
wherein the at least one sample data group comprises at least one positive sample data group and … , there is an interaction event between the sample object and the sample item in the positive sample data group, and there is no interaction event between the sample object and the sample item in the negative sample data group. (see Menon paragraph [0043] "y.sub.i∈{0,1} is a binary label indicating whether a pair is relevant (1) or irrelevant (0). Let n.sub.+ and n.sub.− be the numbers of positively labeled (1) and negatively labeled (0) data points.", and see Menon paragraph [0054] "the label y for any pair … may be a binary indicator of whether the user has "engaged" with the message, where "engaging" may be defined, e.g., as "liking" or replying to a message.")
As to claim 7, Pande-Wang as modified by Menon teaches the method according to claim 5,
wherein the preset loss function comprises a first loss function and a second loss function, the first loss function is a loss function that is set based on an association relationship between the sample object and the sample item, (see Pande paragraph [0046] "a graph-based loss function is applied to the output representations z.sub.u,∀uϵV of the aggregation (step 208). The weight matrices W.sub.k,∀kϵ{1, . . . K} and parameters ... are trained via stochastic gradient descent.", and see Pande paragraph [0047] "r(u,v) is an accumulated mean of the weights on the random walk for node u and v … By adding the weights into the loss function, the algorithm becomes more focused on minimizing the distance between nodes u and v with larger edge weights.", and see Pande paragraph [0051] "the edges of a graph are weighted according to past customer views. Accordingly, weights are provided on all edges, and are calculated based on relative frequency of views for each pair of items … VC(I∩j) is the number of guests that view items i and j in one session")
and the second loss function is a loss function of a preset type. (see Pande paragraph [0046] "L.sub.G(z.sub.u)=−r(u,v)∝log(σ(z.sub.uTz.sub.v))−Q*E.sub.vn˜Pn(v)log(σ(−z.sub.uTz.sub.vn))", and see Pande paragraph [0047] "σ is the sigmoid function, P.sub.n is a negative sampling distribution, and Q defines the number of negative samples.")
As to claim 8, Pande-Wang as modified by Menon teaches the method according to claim 7,
wherein the preset loss function further comprises a third loss function, and the third loss function is a loss function that is set based on an association relationship between at least two sample items. (see Wang paragraph [0040] "the relation matrix for recommended contents specifically refers to a matrix used to represent an association relation between different recommended contents", and see Wang paragraph [0047] " stored in an undirected graph Gr (corresponding to the relation matrix for users in the graph training samples) and an undirected graph Gc (corresponding to the relation matrix for recommended contents in the graph training samples). For each graph (taking Gc as an example, a corresponding processing method is adaptable to Gr), a normalized Laplacian matrix L.sub.c=D.sub.c.sup.−1/2(A.sub.c+I)D.sub.C.sup.−1/2 can be extracted", and see Wang paragraph [0056] "H.sub.Θ.sub.c=GCN(H.sup.0;Θ.sub.c)=relu(L.sub.cH.sup.0Θ.sub.c)", and see Wang paragraph [0061] "the training objective of the low-rank graph convolutional network includes a first parameter item for the first low-rank matrix, a second parameter item for the second low-rank matrix and a non-convex low-rank item", and see Wang paragraph [0078] "r(X)=½(∥W.sub.Θ.sub.r∥.sub.F.sup.2+∥H.sub.Θ.sub.c∥.sub.F.sup.2)−∥W.sub.Θ.sub.rH.sub.Θ.sub.c.sup.T∥.sub.F")
As to claim 9, Pande-Wang as modified by Menon teaches the method according to claim 7,
wherein the preset loss function further comprises a fourth loss function, and the fourth loss function is a loss function that is set based on an association relationship between at least two sample objects. (see Wang paragraph [0037] "The relation matrix for users specifically refers to a matrix used to represent an association relation between different users", and see Wang paragraph [0047] "stored in an undirected graph Gr (corresponding to the relation matrix for users in the graph training samples)", and see Wang paragraph [0056] "W.sub.Θ.sub.r=GCN(W.sup.0;Θ.sub.r)=relu(L.sub.rW.sup.0Θ.sub.r)", and see Wang paragraph [0078] "r(X)=½(∥W.sub.Θ.sub.r∥.sub.F.sup.2+∥H.sub.Θ.sub.c∥.sub.F.sup.2)−∥W.sub.Θ.sub.rH.sub.Θ.sub.c.sup.T∥.sub.F")
As to claim 12, this is directed to an apparatus that corresponds to the method of claim 2, See the rejection for claim 2 above, which also applies to claim 12.
As to claim 14, this is directed to an apparatus that corresponds to the method of claim 4, See the rejection for claim 4 above, which also applies to claim 14.
As to claim 15, this is directed to an apparatus that corresponds to the method of claim 5, See the rejection for claim 5 above, which also applies to claim 15.
As to claim 16, this is directed to an apparatus that corresponds to the method of claim 6, See the rejection for claim 6 above, which also applies to claim 16.
As to claim 17, this is directed to an apparatus that corresponds to the method of claim 7, See the rejection for claim 7 above, which also applies to claim 17.
As to claim 18, this is directed to an apparatus that corresponds to the method of claim 8, See the rejection for claim 8 above, which also applies to claim 18.
As to claim 19, this is directed to an apparatus that corresponds to the method of claim 9, See the rejection for claim 9 above, which also applies to claim 19.
As to claim 22, this is directed to a computer program that corresponds to the method of claim 2, See the rejection for claim 2 above, which also applies to claim 22.
As to claim 24, this is directed to a computer program that corresponds to the method of claim 4, See the rejection for claim 4 above, which also applies to claim 24.
As to claim 25, this is directed to a computer program that corresponds to the method of claim 5, See the rejection for claim 5 above, which also applies to claim 25.
As to claim 26, this is directed to a computer program that corresponds to the method of claim 6, See the rejection for claim 6 above, which also applies to claim 26.
As to claim 27, this is directed to a computer program that corresponds to the method of claim 7, See the rejection for claim 7 above, which also applies to claim 27.
As to claim 28, this is directed to a computer program that corresponds to the method of claim 8, See the rejection for claim 8 above, which also applies to claim 28.
As to claim 29, this is directed to a computer program that corresponds to the method of claim 9, See the rejection for claim 9 above, which also applies to claim 29.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ABDULLAH K ABOUD whose telephone number is (571)272-0025. The examiner can normally be reached Mon-Fri 8am-5pm.
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, Li B Zhen, can be reached at (571) 272-3768. 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.
/ABDULLAH KHALED ABOUD/ Examiner, Art Unit 2121
/Li B. Zhen/ Supervisory Patent Examiner, Art Unit 2121
Prosecution Insights