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 .
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and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114.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 March 4, 2026 has been entered. Accordingly, Claims 1-20 are pending in this application. Claims 1, 10, and 17 are independent claims and have been amended.
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1, 2, 4, and 5 are rejected under 35 U.S.C. 103 as being unpatentable over Larson (PG Pub. No. 2025/0131289 A1), and further in view of Betthauser (PG Pub. No. 2024/0370570 A1).
Regarding Claim 1, Larson discloses a computer-implemented method comprising:
accessing a large language model (LLM) configured to encode textual information to generate encoded textual information, the encoded textual information comprising an embeddings vector representation of the textual information (see Larson, paragraph [0019], where indexing framework 202 enables the extraction of key relationship information; first, the indexing framework 202 embeds text chunks 204 to store in a vector database 206, similar to RAG; the indexing framework 202 simultaneously processes the text chunks using the LLM 104 (via metadata extraction 208) into knowledge graph 106; metadata extraction 208 allows extracting claims, identifying entities of interest 212 and/or establishing relationships 214 between the entities 212);
receiving a data set comprising a plurality of entities and a plurality of relationships among the entities, each relationship associated with respective textual information (see Larson, paragraph [0011], where this system 100 includes a dataset 102 and leverages an LLM 104 to produce an extracted graph representation or knowledge graph (KG) 106 of the dataset 102);
determining a graph representative of the data set, the graph comprising a plurality of nodes and a plurality of edges connecting the nodes, each node representative of a respective entity of the plurality of entities and each edge representative of the relationship between the entities connected by the edge (see Larson, paragraph [0011], where this system 100 includes a dataset 102 and leverages an LLM 104 to produce an extracted graph representation or knowledge graph (KG) 106 of the dataset 102; see also paragraph [0021], where entities and their relationship scan be utilized for knowledge graph induction 306 (e.g., producing knowledge graph 106 that represents entities as nodes/vertices and their relationships as edges)), wherein the determining comprises:
for each edge, applying the LLM to the textual information associated with the relationship represented by the edge to generate encoded edge textual information comprising an embeddings vector representation of the textual information associated with the relationship and adding the encoded edge textual information to the edge in the graph, whereby an enhanced graph is generated (see Larson, paragraph [0019], where indexing framework 202 enables the extraction of key relationship information; first, the indexing framework 202 embeds text chunks 204 to store in a vector database 206, similar to RAG; the indexing framework 202 simultaneously processes the text chunks using the LLM 104 (via metadata extraction 208) into knowledge graph 106; metadata extraction 208 allows extracting claims, identifying entities of interest 212 and/or establishing relationships 214 between the entities 212).
Larson does not disclose training a graph neural network model (GNN) based on the enhanced graph, wherein the GNN receives the embeddings vector representation as features for the edges of the enhanced graph. Larson in view of Betthauser discloses training a graph neural network model (GNN) based on the enhanced graph, wherein the GNN receives the embeddings vector representation as features for the edges of the enhanced graph (see Betthauser, paragraph [0017], where the first model, referred to herein as an encoder model, processes the raw input; the encoder model generates outputs that are used to construct graphs; these graphs are then used to train a second model; see also paragraph [0047], where graph classifier model 530 is one example of a type of graph neural network (GNN) that may be used to infer an output from graphs 400; however, other types of graph neural networks are similarly contemplated).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Larson with Betthauser for the benefit of applying hierarchical ML models to limit the size growth of the LLM (see Betthauser, paragraphs [0016], [0017]).
Regarding Claim 2, Larson in view of Betthauser discloses the computer-implemented method of Claim 1, wherein receiving the LLM comprises training the LLM on a corpus of information specific to a domain represented by the graph (see Larson, paragraph [0010], where retrieval augmented generation (RAG) techniques are the cornerstone of grounding LLMs to domain-specific data).
Regarding Claim 4, Larson in view of Betthauser discloses the computer-implemented method of Claim 1, wherein applying the LLM to the textual information associated with the relationship represented by the edge comprises providing, as input to the LLM: numerical information associated with the relationship represented by the edge (see Larson, paragraph [0018], where the described example case, the internally available dataset covers the Russian invasion of Ukraine … in this implementation, the internally available dataset 102 was created by scraping 97,000 news articles from six news providers, interfaxua, mz, ng, nv, ria, and unian on topics regarding the Russian invasion of Ukraine [it is the position of the Examiner that news articles suggest alphanumeric text, which encompasses numerical information).
Regarding Claim 5, Larson in view of Betthauser discloses the computer-implemented method of Claim 1, wherein applying the LLM to the textual information of an edge comprises providing, as input to the LLM: attributes of nodes connected by the edge (see Larson, paragraph [0019], where indexing framework 202 enables the extraction of key relationship information; first, the indexing framework 202 embeds text chunks 204 to store in a vector database 206, similar to RAG; the indexing framework 202 simultaneously processes the text chunks using the LLM 104 (via metadata extraction 208) into knowledge graph 106; metadata extraction 208 allows extracting claims, identifying entities of interest 212 and/or establishing relationships 214 between the entities 212; see also Fig. 2, where entity nodes have metadata extracted by the LLM [attributes of nodes]).
Claims 3 and 7 are rejected under 35 U.S.C. 103 as being unpatentable over Larson and Betthauser as applied to Claims 1, 2, 4, and 5 above, and further in view of Resheff (PG Pub. No. 2021/0065245 A1).
Regarding Claim 3, Larson in view of Betthauser discloses the computer-implemented method of Claim 1, wherein applying the LLM to the textual information associated with the relationship represented by the edge comprises providing, as input to the LLM:
the textual information associated with the relationship represented by the edge (see Larson, paragraph [0019], where indexing framework 202 enables the extraction of key relationship information; first, the indexing framework 202 embeds text chunks 204 to store in a vector database 206, similar to RAG; the indexing framework 202 simultaneously processes the text chunks using the LLM 104 (via metadata extraction 208) into knowledge graph 106; metadata extraction 208 allows extracting claims, identifying entities of interest 212 and/or establishing relationships 214 between the entities 212).
Larson does not disclose the textual information associated with one or more additional edges that connect the same nodes as the edge. Resheff discloses the textual information associated with one or more additional edges that connect the same nodes as the edge (see Resheff, paragraph [0024], where the relationship graph is a set of two or more nodes connected by one or more edges; a node is an entity within the transactions stored in the data structure 102; an edge is a relationship between two nodes [it is the position of the Examiner that nodes connected by one or more edges, where edges represent relationships, constitutes additional edges that connect the same nodes]).
Larson and Resheff are both directed to relationship graphs and machine learning. Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Larson with Resheff as it amounts to combining prior art elements according to known techniques to yield predictable results (see MPEP 2143(I)(A)).
Regarding Claim 7, Larson in view of Betthauser discloses the computer-implemented method of Claim 1, wherein one or more of:
Larson does not disclose:
one or more entities are end users, and edges representative of relationships between end users comprise textual information comprise user-to-user notes; or
one or more entities are end users and one or more entities are enterprise users, and edges representative of a relationships between an end user and an enterprise user comprise descriptions of resources conveyed from the enterprise user to the end user.
Resheff discloses one or more entities are end users and one or more entities are enterprise users, and edges representative of a relationships between an end user and an enterprise user comprise descriptions of resources conveyed from the enterprise user to the end user (see Resheff, Fig. 7, where edge 3-2 represents a relationship where Node 3 transfers $3000 per month to Node 2 [it is the position of the Examiner that a transfer of $3000 constitutes a conveyance of resources]).
Larson and Resheff are both directed to relationship graphs and machine learning. Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Larson with Resheff as it amounts to combining prior art elements according to known techniques to yield predictable results (see MPEP 2143(I)(A)).
Claims 6 and 8 are rejected under 35 U.S.C. 103 as being unpatentable over Larson and Betthauser as applied to Claims 1, 2, 4, and 5 above, and further in view of Creed (PG Pub. No. 2021/0081717 A1).
Regarding Claim 6, Larson in view of Betthauser discloses the computer-implemented method of Claim 1, further comprising:
Larson does not disclose applying the trained GNN to generate a prediction regarding at least one of the entities. Creed discloses applying the trained GNN to generate a prediction regarding at least one of the entities (see Creed, paragraph [0008], where the trained GNN model may be used to predict link relationship between a first entity and a second entity associated with the entity-entity graph).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Larson with Creed for the benefit of predicting link relationships between entities (see Creed, Abstract).
Regarding Claim 8, Larson in view of Betthauser discloses the computer-implemented method of Claim 1, wherein:
Larson does not disclose training the GNN comprises training the GNN to make a respective predictive classification for each node. Creed discloses training the GNN comprises training the GNN to make a respective predictive classification for each node (see Creed, paragraph [0082], where apparatus and methods describe how to, by way of example only but is not limited to, efficiently increase the robustness of the GCNN technique to generate an accurate GNN model for prediction/classification problems).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Larson with Creed for the benefit of predicting link relationships between entities (see Creed, Abstract).
Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over Larson, Betthauser, and Creed as applied to Claims 6 and 8 above, and further in view of Arora (PG Pub. No. 2022/0101327 A1).
Regarding Claim 9, Larson in view of Betthauser and Creed discloses the computer-implemented of Claim 8, wherein the graph is a first graph, and the method comprising:
Larson does not disclose:
determining a second graph representative of the data set, the second graph comprising the plurality of nodes and the plurality of edges connecting the nodes, wherein determining the second graph comprises:
for each edge of the second graph, applying the LLM to the textual information associated with the relationship represented by the edge and to the predictive classifications of the GNN for the nodes connected by the edge, to generate second encoded edge textual information, whereby a second enhanced graph is generated; and
further training the GNN based on the second enhanced graph.
Larson in view of Arora discloses:
determining a second graph representative of the data set (see Arora, paragraph [0032], where updating the dynamic features further includes modifying one or more existing dynamic features corresponding to the identified cluster due to the formation of new edges in the identified cluster or the modification in the weight of existing edges in the identified cluster; the neural network is re-trained by the server based on the updated dynamic features and one or more other updates in the static features), the second graph comprising the plurality of nodes and the plurality of edges connecting the nodes (see Larson, paragraph [0011], where this system 100 includes a dataset 102 and leverages an LLM 104 to produce an extracted graph representation or knowledge graph (KG) 106 of the dataset 102; see also paragraph [0021], where entities and their relationship scan be utilized for knowledge graph induction 306 (e.g., producing knowledge graph 106 that represents entities as nodes/vertices and their relationships as edges)), wherein determining the second graph comprises:
for each edge of the second graph, applying the LLM to the textual information associated with the relationship represented by the edge (see Larson, paragraph [0011], where this system 100 includes a dataset 102 and leverages an LLM 104 to produce an extracted graph representation or knowledge graph (KG) 106 of the dataset 102; see also paragraph [0021], where entities and their relationship scan be utilized for knowledge graph induction 306 (e.g., producing knowledge graph 106 that represents entities as nodes/vertices and their relationships as edges)), and to the predictive classifications of the GNN for the nodes connected by the edge, to generate second encoded edge textual information, whereby a second enhanced graph is generated (see Arora, paragraph [0032], where using the static and dynamic features, the server trains a neural network for detecting fraudulent transactions … the server further updates the dynamic features including adding new dynamic features corresponding to the new merchant node or the new consumer node added to the identified cluster); and
further training the GNN based on the second enhanced graph (see Arora, paragraph [0032], where updating the dynamic features further includes modifying one or more existing dynamic features corresponding to the identified cluster due to the formation of new edges in the identified cluster or the modification in the weight of existing edges in the identified cluster; the neural network is re-trained by the server based on the updated dynamic features and one or more other updates in the static features).
Larson and Arora are both directed to relationship graphs and machine learning. Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Larson with Arora as it amounts to combining prior art elements according to known techniques to yield predictable results (see MPEP 2143(I)(A)).
Claims 10-12, 14, and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Larson and Betthauser as applied to Claims 1, 2, 4, and 5 above, and further in view of Arora.
Regarding Claim 10, Larson discloses a computing system comprising:
a processor (see Larson, paragraph [0096], where the methods are stored on one or more computer-readable storage medium/media as a set of instructions such that execution by a processor of a computing device causes the computing device to perform the method); and
a non-transitory, computer-readable medium storing instructions that, when executed by the processor, cause the computing system to perform operations (see Larson, paragraph [0096], where the methods are stored on one or more computer-readable storage medium/media as a set of instructions such that execution by a processor of a computing device causes the computing device to perform the method) comprising:
accessing a large language model (LLM) configured to encode textual information to generate encoded textual information, the encoded textual information comprising an embeddings vector representation of the textual information (see Larson, paragraph [0019], where indexing framework 202 enables the extraction of key relationship information; first, the indexing framework 202 embeds text chunks 204 to store in a vector database 206, similar to RAG; the indexing framework 202 simultaneously processes the text chunks using the LLM 104 (via metadata extraction 208) into knowledge graph 106; metadata extraction 208 allows extracting claims, identifying entities of interest 212 and/or establishing relationships 214 between the entities 212);
receiving a data set comprising a plurality of entities and a plurality of relationships among the entities, each relationship associated with respective textual information (see Larson, paragraph [0011], where this system 100 includes a dataset 102 and leverages an LLM 104 to produce an extracted graph representation or knowledge graph (KG) 106 of the dataset 102);
generating a graph comprising a plurality of nodes and a plurality of edges connecting the nodes, each node representative of a respective entity (see Larson, paragraph [0011], where this system 100 includes a dataset 102 and leverages an LLM 104 to produce an extracted graph representation or knowledge graph (KG) 106 of the dataset 102; see also paragraph [0021], where entities and their relationship scan be utilized for knowledge graph induction 306 (e.g., producing knowledge graph 106 that represents entities as nodes/vertices and their relationships as edges)), each edge associated with respective textual information (see Larson, paragraph [0019], where indexing framework 202 enables the extraction of key relationship information; first, the indexing framework 202 embeds text chunks 204 to store in a vector database 206, similar to RAG; the indexing framework 202 simultaneously processes the text chunks using the LLM 104 (via metadata extraction 208) into knowledge graph 106; metadata extraction 208 allows extracting claims, identifying entities of interest 212 and/or establishing relationships 214 between the entities 212), wherein the generating comprises applying the LLM to the textual information associated with each edge to generate encoded edge textual information (see Larson, paragraph [0019], where indexing framework 202 enables the extraction of key relationship information; first, the indexing framework 202 embeds text chunks 204 to store in a vector database 206, similar to RAG; the indexing framework 202 simultaneously processes the text chunks using the LLM 104 (via metadata extraction 208) into knowledge graph 106; metadata extraction 208 allows extracting claims, identifying entities of interest 212 and/or establishing relationships 214 between the entities 212), the encoded edge textual information comprising an embeddings vector representation of the textual information, and adding the encoded edge textual information to the graph (see Larson, paragraph [0011], where this system 100 includes a dataset 102 and leverages an LLM 104 to produce an extracted graph representation or knowledge graph (KG) 106 of the dataset 102; see also paragraph [0021], where entities and their relationship scan be utilized for knowledge graph induction 306 (e.g., producing knowledge graph 106 that represents entities as nodes/vertices and their relationships as edges)), wherein the determining comprises:
Larson does not disclose:
each edge representative of a computing action involving the entities connected by the edge;
training a graph neural network model (GNN) based on the graph, wherein the GNN receives the embeddings vector representation as features for the edges of the graph; and
outputting a prediction about one of the entities according to the trained GNN.
Larson in view of Betthauser discloses training a graph neural network model (GNN) based on the graph, wherein the GNN receives the embeddings vector representation as features for the edges of the graph (see Betthauser, paragraph [0017], where the first model, referred to herein as an encoder model, processes the raw input; the encoder model generates outputs that are used to construct graphs; these graphs are then used to train a second model; see also paragraph [0047], where graph classifier model 530 is one example of a type of graph neural network (GNN) that may be used to infer an output from graphs 400; however, other types of graph neural networks are similarly contemplated).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Larson with Betthauser for the benefit of applying hierarchical ML models to limit the size growth of the LLM (see Betthauser, paragraphs [0016], [0017]).
Larson in view of Betthauser does not disclose:
each edge representative of a computing action involving the entities connected by the edge; and
outputting a prediction about one of the entities according to the trained GNN.
Arora discloses:
each edge representative of a computing action involving the entities connected by the edge (see Arora, paragraph [0123], where each edge belonging to the first edge type ‘A’ between a merchant node and a consumer node implies that at least one approved transaction was executed between a corresponding merchant and a corresponding consumer); and
outputting a prediction about one of the entities according to the trained GNN (see Arora, paragraph [0032], where using the static and dynamic features, the server trains a neural network for detecting fraudulent transactions … the server further updates the dynamic features including adding new dynamic features corresponding to the new merchant node or the new consumer node added to the identified cluster).
Larson and Arora are both directed to relationship graphs and machine learning. Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Larson and Betthauser with Arora as it amounts to combining prior art elements according to known techniques to yield predictable results (see MPEP 2143(I)(A)).
Regarding Claim 11, Larson in view of Betthauser and Arora discloses the computing system of Claim 10, wherein:
Larson does not disclose the prediction comprises a probability of a fraudulent transaction by one of the entities. Arora discloses the prediction comprises a probability of a fraudulent transaction by one of the entities (see Arora, paragraph [0032], where using the static and dynamic features, the server trains a neural network for detecting fraudulent transactions … the server further updates the dynamic features including adding new dynamic features corresponding to the new merchant node or the new consumer node added to the identified cluster).
Larson and Arora are both directed to relationship graphs and machine learning. Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Larson with Arora as it amounts to combining prior art elements according to known techniques to yield predictable results (see MPEP 2143(I)(A)).
Regarding Claim 12, Larson in view of Betthauser and Arora discloses the computing system of Claim 10, wherein applying the LLM to the textual information associated with each edge comprises, for each edge, providing, as input to the LLM, two or more of:
textual information associated with one or more additional edges that connect the same nodes as the edge;
numerical information associated with the relationship represented by the edge (see Larson, paragraph [0018], where the described example case, the internally available dataset covers the Russian invasion of Ukraine … in this implementation, the internally available dataset 102 was created by scraping 97,000 news articles from six news providers, interfaxua, mz, ng, nv, ria, and unian on topics regarding the Russian invasion of Ukraine [it is the position of the Examiner that news articles suggest alphanumeric text, which encompasses numerical information); and
attributes of nodes connected by the edge (see Larson, paragraph [0019], where indexing framework 202 enables the extraction of key relationship information; first, the indexing framework 202 embeds text chunks 204 to store in a vector database 206, similar to RAG; the indexing framework 202 simultaneously processes the text chunks using the LLM 104 (via metadata extraction 208) into knowledge graph 106; metadata extraction 208 allows extracting claims, identifying entities of interest 212 and/or establishing relationships 214 between the entities 212; see also Fig. 2, where entity nodes have metadata extracted by the LLM [attributes of nodes]).
Regarding Claim 14, Larson in view of Betthauser and Arora discloses the computing system of Claim 10, wherein generating the graph further comprises applying the LLM to the textual information associated with each node to generate encoded node textual information and adding the encoded node textual information to the graph (see Larson, paragraph [0019], where indexing framework 202 enables the extraction of key relationship information; first, the indexing framework 202 embeds text chunks 204 to store in a vector database 206, similar to RAG; the indexing framework 202 simultaneously processes the text chunks using the LLM 104 (via metadata extraction 208) into knowledge graph 106; metadata extraction 208 allows extracting claims, identifying entities of interest 212 and/or establishing relationships 214 between the entities 212).
Regarding Claim 16, Larson in view of Betthauser and Arora discloses the computing system of Claim 10, wherein the operations further comprise repeatedly:
Larson does not disclose:
for each edge, applying the LLM to the textual information associated with the computing action represented by the edge, and to respective predictive classifications of the trained GNN for the nodes connected by the edge, to generate second encoded edge textual information and adding the second encoded edge textual information to the edge in the graph, whereby a further enhanced graph is generated;
further training the GNN based on the further enhanced graph; and
re-training the LLM according to predictions made by the further trained GNN.
Larson in view of Arora discloses:
for each edge, applying the LLM to the textual information (see Larson, paragraph [0011], where this system 100 includes a dataset 102 and leverages an LLM 104 to produce an extracted graph representation or knowledge graph (KG) 106 of the dataset 102; see also paragraph [0021], where entities and their relationship scan be utilized for knowledge graph induction 306 (e.g., producing knowledge graph 106 that represents entities as nodes/vertices and their relationships as edges)) associated with the computing action represented by the edge (see Arora, paragraph [0123], where each edge belonging to the first edge type ‘A’ between a merchant node and a consumer node implies that at least one approved transaction was executed between a corresponding merchant and a corresponding consumer), and to respective predictive classifications of the trained GNN for the nodes connected by the edge (see Arora, paragraph [0032], where using the static and dynamic features, the server trains a neural network for detecting fraudulent transactions … the server further updates the dynamic features including adding new dynamic features corresponding to the new merchant node or the new consumer node added to the identified cluster), to generate second encoded edge textual information and adding the second encoded edge textual information to the edge in the graph, whereby a further enhanced graph is generated (see Arora, paragraph [0032], where updating the dynamic features further includes modifying one or more existing dynamic features corresponding to the identified cluster due to the formation of new edges in the identified cluster or the modification in the weight of existing edges in the identified cluster; the neural network is re-trained by the server based on the updated dynamic features and one or more other updates in the static features);
further training the GNN based on the further enhanced graph (see Arora, paragraph [0032], where updating the dynamic features further includes modifying one or more existing dynamic features corresponding to the identified cluster due to the formation of new edges in the identified cluster or the modification in the weight of existing edges in the identified cluster; the neural network is re-trained by the server based on the updated dynamic features and one or more other updates in the static features); and
re-training the LLM according to predictions made by the further trained GNN (see Arora, paragraph [0032], where updating the dynamic features further includes modifying one or more existing dynamic features corresponding to the identified cluster due to the formation of new edges in the identified cluster or the modification in the weight of existing edges in the identified cluster; the neural network is re-trained by the server based on the updated dynamic features and one or more other updates in the static features).
Larson and Arora are both directed to relationship graphs and machine learning. Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Larson with Arora as it amounts to combining prior art elements according to known techniques to yield predictable results (see MPEP 2143(I)(A)).
Claim 13 is rejected under 35 U.S.C. 103 as being unpatentable over Larson, Betthauser, and Arora as applied to Claims 10-12, 14, and 16 above, and further in view of Resheff.
Regarding Claim 13, Larson in view of Betthauser and Arora discloses the computing system of Claim 12, wherein applying the LLM to the textual information associated with each edge comprises, for each edge, providing, as input to the LLM:
numerical information associated with the relationship represented by the edge (see Larson, paragraph [0018], where the described example case, the internally available dataset covers the Russian invasion of Ukraine … in this implementation, the internally available dataset 102 was created by scraping 97,000 news articles from six news providers, interfaxua, mz, ng, nv, ria, and unian on topics regarding the Russian invasion of Ukraine [it is the position of the Examiner that news articles suggest alphanumeric text, which encompasses numerical information); and
attributes of nodes connected by the edge (see Larson, paragraph [0019], where indexing framework 202 enables the extraction of key relationship information; first, the indexing framework 202 embeds text chunks 204 to store in a vector database 206, similar to RAG; the indexing framework 202 simultaneously processes the text chunks using the LLM 104 (via metadata extraction 208) into knowledge graph 106; metadata extraction 208 allows extracting claims, identifying entities of interest 212 and/or establishing relationships 214 between the entities 212; see also Fig. 2, where entity nodes have metadata extracted by the LLM [attributes of nodes]).
Larson does not disclose textual information associated with one or more additional edges that connect the same nodes as the edge. Resheff discloses textual information associated with one or more additional edges that connect the same nodes as the edge (see Resheff, paragraph [0024], where the relationship graph is a set of two or more nodes connected by one or more edges; a node is an entity within the transactions stored in the data structure 102; an edge is a relationship between two nodes [it is the position of the Examiner that nodes connected by one or more edges, where edges represent relationships, constitutes additional edges that connect the same nodes]).
Larson and Resheff are both directed to relationship graphs and machine learning. Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Larson with Resheff as it amounts to combining prior art elements according to known techniques to yield predictable results (see MPEP 2143(I)(A)).
Claim 15 is rejected under 35 U.S.C. 103 as being unpatentable over Larson, Betthauser, and Arora as applied to Claims 10-12, 14, and 16 above, and further in view of Kida (PG Pub. No. 2019/0065989 A1).
Regarding Claim 15, Larson in view of Betthauser and Arora discloses the computing system of Claim 10, wherein the operations further comprise:
Larson does not disclose training the LLM using a training data set comprising computing actions involving a subset of the entities represented in the graph. Kida discloses training the LLM using a training data set comprising computing actions involving a subset of the entities represented in the graph (see Kida, paragraph [0071], where example 1 is a system for selecting training set samples).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Larson with Kida as it amounts to simple substitution of one known element for another to obtain a predictable result (see MPEP 2143(I)(B)).
Claims 17, 18, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Larson and Betthauser as applied to Claims 1, 2, 4, and 5 above, and further in view of Liu (“Explore then Determine: A GNN-LLM Synergy Framework for Reasoning over Knowledge Graph”, Guangyu Liu, Tsinghua University, June 3, 2024).
Regarding Claim 17, Larson discloses a computer-implemented method comprising:
accessing a large language model (LLM) configured to encode textual information to generate encoded textual information, the encoded textual information comprising an embeddings vector representation of the textual information (see Larson, paragraph [0019], where indexing framework 202 enables the extraction of key relationship information; first, the indexing framework 202 embeds text chunks 204 to store in a vector database 206, similar to RAG; the indexing framework 202 simultaneously processes the text chunks using the LLM 104 (via metadata extraction 208) into knowledge graph 106; metadata extraction 208 allows extracting claims, identifying entities of interest 212 and/or establishing relationships 214 between the entities 212);
accessing a graph comprising a plurality of nodes and a plurality of edges connecting the nodes, each node representative of a respective entity and each edge representative of a relationship between the entities connected by the edge, each edge associated with respective textual information (see Larson, paragraph [0019], where indexing framework 202 enables the extraction of key relationship information; first, the indexing framework 202 embeds text chunks 204 to store in a vector database 206, similar to RAG; the indexing framework 202 simultaneously processes the text chunks using the LLM 104 (via metadata extraction 208) into knowledge graph 106; metadata extraction 208 allows extracting claims, identifying entities of interest 212 and/or establishing relationships 214 between the entities 212); and
applying the LLM to the textual information associated with each edge to generate encoded edge textual information, the encode edge textual information comprising an embeddings vector representation of the textual information associated with the edge, and adding the encoded edge textual information to the graph to yield an enhanced graph (see Larson, paragraph [0019], where indexing framework 202 enables the extraction of key relationship information; first, the indexing framework 202 embeds text chunks 204 to store in a vector database 206, similar to RAG; the indexing framework 202 simultaneously processes the text chunks using the LLM 104 (via metadata extraction 208) into knowledge graph 106; metadata extraction 208 allows extracting claims, identifying entities of interest 212 and/or establishing relationships 214 between the entities 212).
Larson does not disclose:
training a graph neural network model (GNN) based on the enhanced graph, wherein the GNN receives the embeddings vector representation as features for the edges of the enhanced graph;
training the LLM according to node predictions made by the trained GNN;
re-applying the trained LLM to the textual information associated with each edge to generate further encoded edge textual information to yield a further enhanced graph; and
further training the GNN based on the further enhanced graph.
Betthauser discloses training a graph neural network model (GNN) based on the enhanced graph, wherein the GNN receives the embeddings vector representation as features for the edges of the enhanced graph (see Betthauser, paragraph [0017], where the first model, referred to herein as an encoder model, processes the raw input; the encoder model generates outputs that are used to construct graphs; these graphs are then used to train a second model; see also paragraph [0047], where graph classifier model 530 is one example of a type of graph neural network (GNN) that may be used to infer an output from graphs 400; however, other types of graph neural networks are similarly contemplated).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Larson with Betthauser for the benefit of applying hierarchical ML models to limit the size growth of the LLM (see Betthauser, paragraphs [0016], [0017]).
Larson in view of Betthauser does not disclose:
training the LLM according to node predictions made by the trained GNN;
re-applying the trained LLM to the textual information associated with each edge to generate further encoded edge textual information to yield a further enhanced graph; and
further training the GNN based on the further enhanced graph.
Larson in view of Betthauser and Liu discloses:
training the LLM according to node predictions (see Liu, Fig. 1, Illustration of the Explore-then-Determine (EtD) framework that synergizes LLMs with GNNs for reasoning over KG);
re-applying the trained LLM to the textual information associated with each edge to generate further encoded edge textual information to yield a further enhanced graph (see Larson, paragraph [0019], where indexing framework 202 enables the extraction of key relationship information; first, the indexing framework 202 embeds text chunks 204 to store in a vector database 206, similar to RAG; the indexing framework 202 simultaneously processes the text chunks using the LLM 104 (via metadata extraction 208) into knowledge graph 106; metadata extraction 208 allows extracting claims, identifying entities of interest 212 and/or establishing relationships 214 between the entities 212); and
further training the GNN based on the further enhanced graph (see Betthauser, paragraph [0017], where the first model, referred to herein as an encoder model, processes the raw input; the encoder model generates outputs that are used to construct graphs; these graphs are then used to train a second model; see also paragraph [0047], where graph classifier model 530 is one example of a type of graph neural network (GNN) that may be used to infer an output from graphs 400; however, other types of graph neural networks are similarly contemplated).
Larson, Betthauser, and Liu are all directed toward LLMs, GNNs, and Knowledge Graphs. Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Larson and Betthauser with Liu as it amounts to combining prior art elements according to known techniques to yield predictable results (see MPEP 2143(I)(A)).
Regarding Claim 18, Larson in view of Betthauser and Liu discloses the computer-implemented method of Claim 17, further comprising:
Larson does not disclose repeating the training the LLM, re-applying the trained LLM, and further training the GNN, and deploying the trained GNN after the repeating. Liu discloses repeating the training the LLM, re-applying the trained LLM, and further training the GNN, and deploying the trained GNN after the repeating (see Liu, Fig. 1, Illustration of the Explore-then-Determine (EtD) framework that synergizes LLMs with GNNs for reasoning over KG [it is the position of the Examiner that the continuous symbiotic training of LLM and GNN contemplates the re-training and re-applying steps claimed by the Applicant]).
Larson, Betthauser, and Liu are all directed toward LLMs, GNNs, and Knowledge Graphs. Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Larson with Liu as it amounts to combining prior art elements according to known techniques to yield predictable results (see MPEP 2143(I)(A)).
Regarding Claim 20, Larson in view of Betthauser and Liu discloses the computer-implemented method of Claim 17, wherein applying the LLM to the textual information of an edge comprises providing, as input to the LLM: attributes of nodes connected by the edge (see Larson, paragraph [0019], where indexing framework 202 enables the extraction of key relationship information; first, the indexing framework 202 embeds text chunks 204 to store in a vector database 206, similar to RAG; the indexing framework 202 simultaneously processes the text chunks using the LLM 104 (via metadata extraction 208) into knowledge graph 106; metadata extraction 208 allows extracting claims, identifying entities of interest 212 and/or establishing relationships 214 between the entities 212; see also Fig. 2, where entity nodes have metadata extracted by the LLM [attributes of nodes]).
Claim 19 is rejected under 35 U.S.C. 103 as being unpatentable over Larson, Betthauser, and Liu, as applied to Claims 17, 18, and 20 above, and further in view of Arora.
Regarding Claim 19, Larson in view of Betthauser and Liu discloses the computer-implemented method of Claim 17, further comprising:
Larson does not disclose:
receiving a request from a user represented in the graph to perform a computing action;
determining a fraud risk of the computing action according to the trained GNN; and
generating a response to the user according to the fraud risk.
Arora discloses:
receiving a request from a user represented in the graph to perform a computing action (see Arora, paragraph [0032], where new transaction data is received for a new transaction, the server detects the new transaction as one of a fraudulent transaction or legitimate transaction based on an output of the trained neural network for the inputted new transaction data);
determining a fraud risk of the computing action according to the trained GNN (see Arora, paragraph [0032], where new transaction data is received for a new transaction, the server detects the new transaction as one of a fraudulent transaction or legitimate transaction based on an output of the trained neural network for the inputted new transaction data); and
generating a response to the user according to the fraud risk (see Arora, paragraph [0047], where in one embodiment, the payment network server 106 is configured to decline a transaction that is detected as fraudulent).
Larson and Arora are both directed to relationship graphs and machine learning. Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to combine Larson with Arora as it amounts to combining prior art elements according to known techniques to yield predictable results (see MPEP 2143(I)(A)).
Response to Arguments
Applicant’s Arguments, filed March 4, 2026 have been fully considered, but they are not persuasive in view of the new grounds of rejection.
Conclusion
The prior art made of record and not relied upon is considered pertinent to the Applicant’s disclosure:
Zadeh (PG Pub. No. 2020/0287927 A1), which concerns anomaly detection based changes to an entity relationship graph.
Oberbreckling (PG Pub. No. 2018/0075104 A1), which concerns relationship discovery between datasets.
Bratanic (Constructing knowledge graphs from text using OpenAI functions, http://www.bratanic-tomaz.medium.com/constructing-knowledge-graphs-from-text-using-openai-functions-096a6d010c17, October 20, 2023), which concerns constructing knowledge graphs using LLMs in Neo4j.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to FARHAD AGHARAHIMI whose telephone number is (571)272-9864. The examiner can normally be reached M-F 9am - 5pm ET.
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, Apu Mofiz can be reached at 571-272-4080. 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.
/FARHAD AGHARAHIMI/Examiner, Art Unit 2161
/APU M MOFIZ/Supervisory Patent Examiner, Art Unit 2161
Prosecution Insights