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 .
Status of the Claims
Claims 1-20 are pending for examination.
Claims 1, 10, 19 are independent Claims.
Claims 19-20 are rejected under 35 U.S.C. §101.
Claims 1-20 are rejected under 35 U.S.C. §103.
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 19-20 are rejected under 35 U.S.C. 101 because the claim is directed to “a computer product”. The computer product is not limited by Applicant’s Specification. By broadest reasonable interpretation, “a computer product” is construed as transitory signal for storing program instruction. The Claims are rejected for being Signal per Se.
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) 1-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lee et al. (U.S. 2020/0285944 hereinafter Lee) in view of Larson et al. (U.S. 2025/0328565 hereinafter Larson).
As Claim 1, Lee teaches a method comprising:
constructing a graph neural network (Lee (¶0056 line 1-5, fig. 2 item 220), “the one or more computing systems perform a training process to train a graph convolutional network that includes two or more graph convolutional layers, for example, using a portion of the graph-structure dataset.”) from an input graph comprising a plurality of nodes and one or more edges (Lee (¶0055 line 1-6, fig. 2 item 210), “the one or more computing systems receive a graph-structured dataset that identifies a set of entities ( e.g., sessions of web-based services or users of social networks), features of the entities ( e.g., attributes of the sessions or users), and connections between the entities (e.g., relations or links between the sessions or users).”), the graph neural network comprising one or more internal layers (Lee (¶0056 line 1-5, fig. 2 item 220), “two or more graph convolutional layers”), each internal layer comprising one or more node vectors encoding a K-hop neighborhood for a target node of the plurality of nodes (Lee (¶0059 line 1-7), “various motifs to select nodes within different neighborhoods for information integration at different target nodes in the graph, where the different neighborhoods include nodes within different distances from respective target nodes and/or nodes forming different structural relationships (i.e., motifs) with the respective target nodes”);
generating, from the graph neural network, a representation for the target node (Lee (¶0075 last 4 lines), “the feature representation of target node 426 includes the aggregation (e.g., weighted sum) of the original features of nodes 408, 412, 414, 416, 418, 422, 424, and 426”); and
generating, from a feed forward neural network tower (Lee (¶0041 line 9-10), feed-forward neural network), an output vector (Lee (¶0063 line 4-5), “NxF1 an feature matrix where each row is a feature representation of a node”) for the target node using the representation for the target node (Lee (¶0075 last 4 lines), “the feature representation of target node 426 includes the aggregation (e.g., weighted sum) of the original features of nodes 408, 412, 414, 416, 418, 422, 424, and 426”).
Lee may not explicitly disclose:
receiving retrieval augmented generation (RAG) data comprising non-graph contextual data for each of the plurality of nodes;
generating, by a large language model encoder, RAG embeddings from the non-graph contextual data;
encoding the RAG embeddings for each node of the plurality of nodes into the respective node vectors;
Larson teaches:
receiving retrieval augmented generation (RAG) data comprising non-graph contextual data for each of the plurality of nodes (Larson (¶0086 line 1-2), “when given a technical document that is desired to be used as a RAG reference”);
generating, by a large language model encoder, RAG embeddings from the non-graph contextual data (Larson (¶0086 line 3-7), “constructing (e.g., via execution of a text-to graph neural network, or via leveraging of an LLM with a set of document-exploration prompts) a document-graph to represent the substantive or semantic content and structure of that given technical document”);
encoding the RAG embeddings for each node of the plurality of nodes into the respective node vectors (Larson (¶0129 line 1-2), “the context-tagged text block 506(1)(p1) can correspond to an embedding 508(1) (p1)”) of the graph neural network (Larson (¶0086 last 7 lines), “each text block of the given technical document can be considered as being tagged with whatever contextual information that applies to it”);
Lee discloses a system/method for generating node representation based on neighbor nodes. Larson discloses a system/method for generating context-tag for each node in a graph using RAG method. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify nodes of Lee instead be a context tagged nodes taught by Larson, with a reasonable expectation of success. The motivation would be to “to avoid situations in which RAG-AI mistakenly or erroneously synthesizes answers for certain scientific instruments using text blocks that are not applicable or pertinent to those certain scientific instruments.” (Larson (¶0037 last 7 lines))
As Claim 2, besides Claim 1, Lee in view of Larson teaches wherein K is two (Lee (¶0074 line 5-10), “two-hop neighbors”) and the graph neural network comprises three internal layers (Lee (¶0073 line 1-3), “FIG. 4 illustrates an example of a process 400 for information integration at different layers in an example of a deep graph convolutional network”), the three internal layers comprising a first internal layer encoding a 2-hop neighborhood for the target node (Lee (¶0074 line 5-10), “Because features of nodes 418, 422, and 424 generated at first GCN layer 410 include the original features of nodes 414 and 416, the new features for target node 426 generated by second GCN layer 420 now include the original features of nodes 414 and 416 as well, which are the
two-hop neighbors of target node”), a second internal layer encoding a 1-hop neighborhood for the target node (Lee (¶0073 line 7-11), “At a first GCN layer 410, original features of nodes 418, 422, 424, and 426, which are the one-hop neighbors of target node 426 (including target node 426 itself), are aggregated (e.g., as a weighted sum) to generate new features for target node 426, as shown by the dash lines”), and a third internal layer encoding a 2-hop neighborhood for the target node (Lee (¶0074 last 11 lines), “Because the features of nodes 422 and 418 generated by second GCN layer 420 include the original features of nodes 408 and 412, which are the three-hop neighbors of target node 426, the new features for target node 426 generated by third GCN layer 430 now includes the original features of the three-hop neighbors (such as nodes 408 and 412) of target node 426.”).
As Claim 3, besides Claim 1, Lee in view of Larson teaches further comprising coupling the output vector to a loss function with a second output vector from a second feed forward neural network tower (Lee (¶0105 line 1-5), “While the cross-entropy loss function described in equation (21) is sufficient for training the MCN to classify nodes in input graphs, it may not be sufficient for training the attention mechanism that selects the best motif and step-size for each node at each layer.”).
As Claim 4, besides Claim 3, Lee in view of Larson teaches wherein the second feed forward neural network tower is coupled to a second large language model encoder, and wherein an input to the second feed forward neural network tower comprises an embedding, from the second large language model, of a query comprising textual data (Larson (¶0128 line 9-13), “forward pass through the one or more hidden layers of the encoder portion 308, and the output layer of the encoder portion 308 can compute the embedding 508(1) based on whatever activation maps or feature maps are provided by its one or more hidden layers”).
As Claim 5, besides Claim 3, Lee in view of Larson teaches wherein the second feed forward neural network tower is coupled to one or more of a convolutional neural network (CNN) (Lee (¶0041 line 1-3, 7 and 9-10), “convolutional neural networks (CNNs)” and “fee-forward neural networks”) or a vision Transformer (ViT), and wherein an input to the second feed forward neural network tower comprises an embedding, from one of the CNN or the ViT, of image data (Larson (¶0128 line 6-13), “the search component 320 can feed the context-tagged text block 506(1)(1) to the input layer of the encoder portion 308, the context-tagged text block 506(1)(1) can complete a forward pass through the one or more hidden layers of the encoder portion 308, and the output layer of the encoder portion 308 can compute the embedding 508(1) based on whatever activation maps or feature maps are provided by its one or more hidden layers”).
As Claim 6, besides Claim 3, Lee in view of Larson teaches further comprising selecting, via a gate selection module, the second output vector from the second feed forward neural network tower from a plurality of additional feed forward neural network towers having outputs coupled to the gate selection module (Larson (¶0088 line 1-7), “when given a set of potentially-relevant context-tagged text blocks (e.g., that have been returned via embedding searches, keyword searches, or text block com positions), a re-ranker can be implemented to identify which of those potentially-relevant context-tagged text blocks are more relevant to a given question”).
As Claim 7, besides Claim 3, Lee in view of Larson teaches further comprising:
storing a plurality of output vectors for a plurality of target nodes in a first embedding database (Larson (¶0162 line 2-5), “the search component 320 can execute the encoder portion 308 on the plain text question 304, thereby yielding an embedding (e.g., a latent vector representation) for the plain text question 304”); and
storing a plurality of second output vectors from the second feed forward neural network tower in a second embedding database (Larson (¶0162 line 5-10), “the search component 320 can search for context-tagged text blocks in the document-graph repository 402 whose embeddings are within any suitable threshold margin of similarity ( e.g., in terms of Euclidean distance or cosine similarity) of the embedding of the plain text question 304”).
As Claim 8, besides Claim 7, Lee in view of Larson teaches further comprising:
receiving a request from a user (Larson (¶0162 line 2-5), “the search component 320 can execute the encoder portion 308 on the plain text question 304, thereby yielding an embedding (e.g., a latent vector representation) for the plain text question 304”); and
providing, responsive to the request, a recommendation to the user, wherein the recommendation comprises a new edge in the graph neural network (Larson (¶0162 last 5 lines), “any context-tagged text block whose embedding differs from that of the plain text question 304 by less than that threshold margin of similarity can be considered or otherwise deemed to be potentially-relevant to the plain text question”).
As Claim 9, besides Claim 8, Lee in view of Larson teaches further comprising:
retrieving an embedding from the second embedding database corresponding to the request (Larson (¶0162 line 2-5), “the search component 320 can execute the encoder portion 308 on the plain text question 304, thereby yielding an embedding (e.g., a latent vector representation) for the plain text question 304”);
determining a set of K-nearest members for the retrieved embedding from the second embedding database (Larson (¶0162 line 5-10), “the search component 320 can search for context-tagged text blocks in the document-graph repository 402 whose embeddings are within any suitable threshold margin of similarity ( e.g., in terms of Euclidean distance or cosine similarity) of the embedding of the plain text question 304”); and
retrieving an embedding from the first embedding database corresponding to each member of the set of K-nearest members (Larson (¶0162 line 5-10), “the search component 320 can search for context-tagged text blocks in the document-graph repository 402 whose embeddings are within any suitable threshold margin of similarity ( e.g., in terms of Euclidean distance or cosine similarity) of the embedding of the plain text question 304”);
wherein the recommendation is generated based on the respective retrieved embeddings from the first embedding database and the second embedding database (Larson (¶0162 last 5 lines), “any context-tagged text block whose embedding differs from that of the plain text question 304 by less than that threshold margin of similarity can be considered or otherwise deemed to be potentially-relevant to the plain text question”).
As Claim 10, Lee teaches a system having a memory, computer readable instructions, and one or more processors for executing the computer readable instructions, the computer readable instructions controlling the one or more processors to perform operations (Lee (¶0147 line 6-8), “a computing system 1500 includes a processor 1502 communicatively coupled to one or more memory devices 1504.”) comprising:
The rest of the limitations are rejected for the same reasons as Claim 1.
As Claims 11-18, the Claims are rejected for the same reasons as Claims 2-9, respectively.
As Claims 19, Lee teaches a computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by one or more processors to cause the one or more processors to perform operations (Lee (¶0147 line 6-8), “a computing system 1500 includes a processor 1502 communicatively coupled to one or more memory devices 1504.”) comprising:.
The rest of the limitations are rejected for the same reasons as Claim 1.
As Claim 20, the Claim is rejected for the same reasons as Claim 2.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Xu et al. (U.S. 2020/0104366) teaches a system/method for generating node embeddings using graph-to-sequence model.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to NHAT HUY T NGUYEN whose telephone number is (571)270-7333. The examiner can normally be reached M-F: 12:00-8:00 EST.
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, Viker Lamardo can be reached at 571-270-5871. 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.
/NHAT HUY T NGUYEN/ Primary Examiner, Art Unit 2147
Prosecution Insights