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 .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 2025/02/21. The submission is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Claim Objections
Claim 20 is objected to because of the following informality: claim 20 recites “the computer system of claim 18,” but claim 18 is a non-transitory computer readable storage medium claim and not a computer system claim. This appears to be a typographical error, as the only computer system claim from which claim 20 can properly depend is claim 19. For purposes of this examination, the examiner treats claim 20 as depending from computer system claim 19, and claim 20 has been examined accordingly. Appropriate correction is required.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1–20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding independent claims 1, 10, and 19
Step 1 — whether the claim falls within a statutory category. See MPEP 2106.03.
Claim 1 is drawn to a method claim; claim 10 is drawn to a non-transitory computer readable storage medium claim; claim 19 is drawn to a system claim. Therefore, each of these claims falls under one of the four categories of statutory subject matter (process/method, machine/product/apparatus, manufacture, or composition of matter). The analysis proceeds to Step 2A.
Step 2A Prong One — whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
Regarding independent claim 1, the claim recites the limitations of "obtaining descriptions for an item," "specifying a request to extract a set of attributes for the item from the description of the item," "extract the set of attributes for the item," "for each attribute, generating connections between an item node representing the item and a set of attribute nodes for the extracted set of attributes," "identifying one or more relevant nodes ... for the user," and "identify one or more item nodes connected to the relevant nodes."
These limitations, under their broadest reasonable interpretation, are directed to the abstract idea of a mental process, namely obtaining item information, extracting attributes of the item, associating the item with its attributes, and identifying items related to a user, which are concepts that can be performed in the human mind, or by a human with the aid of pen and paper, as observation, evaluation, and judgment (see MPEP 2106.04(a)(2), subsection III). For example, a person, such as a store clerk, can read the description or label of an item, note the item's attributes, mentally or on paper associate the item with those attributes and with other items sharing the attributes, and identify related items for a customer.
The claim further recites the limitation of "presenting items represented by the one or more item nodes as recommendations to the user." This limitation is directed to the abstract idea of a certain method of organizing human activity, specifically advertising, marketing, or sales activities or behaviors (see MPEP 2106.04(a)(2), subsection II), as recommending items to a user is a marketing or sales activity.
Independent claim 10 is a non-transitory computer readable storage medium claim reciting similar limitations to claim 1 and is directed to the abstract idea for similar reasons. Independent claim 19 is a system claim reciting similar limitations to claim 1 and is directed to the abstract idea for similar reasons. Accordingly, claims 1, 10, and 19 recite a judicial exception, and the analysis proceeds to Step 2A Prong Two.
Step 2A Prong Two — whether the claim as a whole integrates the recited judicial exception into a practical application. This evaluation is performed by (1) identifying whether there are any additional elements recited in the claim beyond the judicial exception, and (2) evaluating those additional elements individually and in combination to determine whether the claim as a whole integrates the exception into a practical application. See MPEP 2106.04(d).
Regarding independent claim 1, this claim recites the additional elements of "a knowledge graph database," "a machine-learned language model," "one or more prompts for input to a machine-learned language model," "a model serving system," "an item node," "a set of attribute nodes," "an online system," and "traversing the knowledge graph database."
Evaluating these additional elements individually and in combination:
The recited "machine-learned language model," "model serving system," "knowledge graph database," "item node," and "attribute nodes" amount to no more than mere instructions to apply the abstract idea using generic computer components that merely act as tools on which the abstract idea is performed (see MPEP 2106.05(f)). The claim recites generating prompts for, and executing, a machine-learned language model, and storing data as nodes and connections in a database, at a high level of generality, without reciting any specific improvement to the machine-learned language model, to the database, or to the functioning of a computer (see MPEP 2106.05(a)). Any asserted improvement resides in the accuracy or relevance of the item recommendations, which is an improvement to the abstract idea itself rather than a technical improvement to a computer or other technology.
The limitation of "obtaining descriptions for an item" is insignificant extra-solution activity, namely mere data gathering, and the limitation of "presenting items ... as recommendations to the user" is insignificant extra-solution activity, namely outputting or displaying a result (see MPEP 2106.05(g)).
The recited "online system," and the general environment of a knowledge graph database, amount to no more than generally linking the use of the judicial exception to a particular technological environment or field of use (see MPEP 2106.05(h)).
These additional elements, considered individually and in combination, do not impose any meaningful limits on the abstract idea and merely use generic computer components as tools to implement the abstract idea. Therefore, claim 1 as a whole does not integrate the recited judicial exception into a practical application, and claim 1 is directed to the abstract idea.
Regarding independent claim 10, this claim is drawn to a non-transitory computer readable storage medium claim reciting similar limitations to claim 1 and is rejected under the same rationale. Claim 10 also recites the additional elements of "a non-transitory computer readable storage medium comprising stored program code instructions" and "a processing system." These limitations amount to no more than mere instructions to apply the exception using a generic computer or generic computer components that merely act as a tool on which the abstract idea operates (see MPEP 2106.05(f)), and thus fail to integrate the exception into a practical application.
Regarding independent claim 19, this claim is drawn to a system claim reciting similar limitations to claim 1 and is rejected under the same rationale. Claim 19 also recites the additional elements of "a processor" and "a non-transitory computer-readable medium storing instructions." These limitations amount to no more than mere instructions to apply the exception using a generic computer or generic computer components that merely act as a tool on which the abstract idea operates (see MPEP 2106.05(f)), and thus fail to integrate the exception into a practical application.
Step 2B — whether the claim provides an inventive concept, i.e., whether the additional elements, individually and in combination, amount to significantly more than the judicial exception. See MPEP 2106.05.
Regarding independent claims 1, 10, and 19, the additional elements identified above—the machine-learned language model, model serving system, knowledge graph database, item node, attribute nodes, processing system, processor, non-transitory computer readable medium, and online system—are recited at a high level of generality and perform generic computer functions of executing a model, storing and organizing data, parsing data, and traversing stored data. These are well-understood, routine, and conventional computer functions previously known to the industry (see MPEP 2106.05(d)). Executing a machine-learned language model on a prompt to obtain a response, storing data in a graph database as nodes and connections, and parsing a response are generic computer operations recited without any specific inventive implementation.
As addressed in Step 2A Prong Two, the additional elements of "obtaining descriptions for an item" and "presenting items ... as recommendations to the user" constitute insignificant extra-solution activity, namely data gathering and outputting a result, which cannot provide an inventive concept (see MPEP 2106.05(g)).
Considered individually and as an ordered combination, the additional elements do not amount to significantly more than the abstract idea; they merely apply the abstract idea using generic computer components. Therefore, claims 1, 10, and 19 do not include additional elements that are sufficient to amount to significantly more than the judicial exception, and claims 1, 10, and 19 are not patent eligible.
Regarding dependent claims 2–9, 11–18, and 20
Step 1 — See MPEP 2106.03. Claims 2–9 are drawn to method claims; claims 11–18 are drawn to non-transitory computer readable storage medium claims; claim 20 is drawn to a system claim. Each falls within a statutory category.
Step 2A Prong One — See MPEP 2106.04.
Regarding claims 2, 11, and 20, these claims recite the limitation of extracting "a characteristic of the item, a health-related concept of the item, a brand of the item, or a taxonomical category of the item." These limitations further describe the abstract idea of a mental process of extracting and characterizing item attributes, a concept that can be performed in the human mind as observation, evaluation, and judgment.
Regarding claims 3 and 12, these claims recite "identifying a subset of nodes having embeddings less than a threshold distance from an embedding of the relevant node" and generating and comparing embeddings. These limitations are directed to the abstract idea of mathematical concepts, namely mathematical relationships and calculations (see MPEP 2106.04(a)(2), subsection I), and to a mental process of identifying related items by comparison, and further recite the certain-method-of-organizing-human-activity of presenting recommendations.
Regarding claims 4 and 13, these claims recite that "an embedding for a node incorporates semantic information for text describing the node." This limitation is directed to the abstract idea of mathematical concepts, namely a mathematical representation of information.
Regarding claims 5 and 14, these claims recite "applying parameters to embeddings," "combining the set of intermediate embeddings to generate an updated embedding," "computing a loss function based on the updated embedding," and "backpropagating error terms from the loss function to update the parameters." These limitations are directed to the abstract idea of mathematical concepts, namely mathematical calculations and mathematical relationships (see MPEP 2106.04(a)(2), subsection I).
Regarding claims 8 and 17, these claims recite "specifying another request to infer a connection between a first entity ... and a second entity" and "identify whether the connection exists between the first node and the second node." These limitations further describe the abstract idea of a mental process of evaluating and judging whether a relationship exists between two items.
Regarding claims 9 and 18, these claims recite "receiving feedback from the user on the presented items" and "performing a retraining process to update the embeddings." These limitations are directed to the abstract idea of a mental process of evaluating feedback and to mathematical concepts of recomputing the embeddings.
Claims 6, 7, 15, and 16 merely narrow the previously cited abstract idea by specifying a data structure ("an array of lists," claims 6 and 15) for storing the organized information and a source of the item descriptions ("a label of the item or instructions for the item provided by a retailer," claims 7 and 16), and do not recite additional judicial exceptions beyond those of the independent claims.
Step 2A Prong Two and Step 2B — See MPEP 2106.04(d) and 2106.05.
Regarding claims 2–9, 11–18, and 20, these claims recite the additional elements previously identified with respect to the independent claims, including the machine-learned language model, knowledge graph database, embeddings, nodes, processing system, and processor. As with the independent claims, these additional elements amount to no more than mere instructions to apply the abstract idea using generic computer components that merely act as tools on which the abstract idea operates (see MPEP 2106.05(f)), generally linking the use of the exception to a particular technological environment (see MPEP 2106.05(h)), and performing well-understood, routine, and conventional computer functions (see MPEP 2106.05(d)). The recitation of storing the graph as "an array of lists" (claims 6, 15) recites a generic and conventional data structure, and "obtaining the descriptions from a label of the item or instructions for the item provided by a retailer" (claims 7, 16) is insignificant extra-solution data-gathering activity (see MPEP 2106.05(g)).
For the reasons described above with respect to independent claims 1, 10, and 19, the dependent claims merely narrow the recited abstract ideas and do not meaningfully integrate the judicial exceptions into a practical application or amount to significantly more than the judicial exceptions. Therefore, claims 2–9, 11–18, and 20 are also directed to abstract ideas without significantly more and are rejected under 35 U.S.C. 101.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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 text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
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.
Claims 1, 2, 8, 10, 11, 17, 19, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Yu et al. (Yu), Non-Patent Literature, "FolkScope: Intention Knowledge Graph Construction for E-commerce Commonsense Discovery," arXiv:2211.08316v1, published November 15, 2022, in view of Moon et al. (Moon), U.S. Patent No. 11,442,992 B1, cited in the IDS filed on 2/21/25.
Regarding claim 1, (Yu) teaches a method comprising:
[A] "generating a knowledge graph database representing relationships between entities of an online system, generating the knowledge graph database comprising:" (Yu) teaches this limitation. (Yu) constructs an intention knowledge graph for an e-commerce platform (the online system) and populates it with 184,146 item entities, 217,108 intention entities, and 12,755,525 edges representing relationships (assertions) between the entities ((Yu), page 3, § 2 contributions; Figure 1, page 2, depicting item and concept entities linked by IsA, HasProperty, UsedFor, and CapableOf relation edges).
[B] "obtaining descriptions for an item," (Yu) teaches this limitation. (Yu) obtains item titles (descriptions) from the Amazon Review Data (2018) corpus and verbalizes prompt templates using the titles of the item pairs ((Yu), page 4, § 3.2 "User Behavior Data Sampling," and § 3.3 "Knowledge Generation").
[C] "generating one or more prompts for input to a machine-learned language model, the one or more prompts specifying a request to extract a set of attributes for the item from the description of the item," (Yu) teaches this limitation. (Yu) designs a prompt of the form "A user bought item 1 and item 2 because [GEN]" for input to a large language model, and aligns the prompt templates with eighteen ConceptNet relations to request extraction of the item's attributes/relations ((Yu), page 2, § 1; page 4, § 3.3 and Table 2 "Prompts for different commonsense relations"; page 3, § 2 "Language Models as Knowledge Bases," describing designing prompts to probe knowledge from the language model).
[D] "receiving, from a model serving system, responses generated by executing the machine-learned language model on the one or more prompts," (Yu) teaches this limitation. (Yu) executes a pretrained language model on the prompt and receives the generated candidate assertions in return, as shown by the Prompt → Pretrained Language Model → Generate pipeline ((Yu), page 3, Figure 2; page 2, § 1, stating the language model generates candidates in response to the prompt).
[E] "parsing the response from the model serving system to extract the set of attributes for the item, and" (Yu) teaches this limitation. (Yu) structures the generated responses by performing pattern mining on their dependency parses to aggregate the assertions, and then filters and populates the extracted attribute assertions ((Yu), page 4, § 3.4; page 5, § 3.4.2 "Population," retaining the plausible assertions).
[F] "for each attribute, generating connections between an item node representing the item and a set of attribute nodes for the extracted set of attributes in the database;" (Yu) teaches this limitation. (Yu) connects each item node to a set of intention/property/concept (attribute) nodes by edges (assertions), forming the 12,755,525 edges of the knowledge graph ((Yu), Figure 1, page 2, showing an item node connected by HasProperty, UsedFor, and IsA edges to concept nodes; page 3, § 2 contributions).
(Yu) teaches subject matter related to limitations [G], [H], and [I] — namely, incorporating the constructed knowledge graph into a recommendation task by learning item embeddings from the graph to recommend items to users ((Yu), page 8, § Recommendation). However, (Yu) does not teach:
[G] "responsive to receiving an indication that a user accesses the online system, identifying one or more relevant nodes in the knowledge graph database for the user;"
[H] "traversing the knowledge graph database to identify one or more item nodes connected to the relevant nodes; and"
[I] "presenting items represented by the one or more item nodes as recommendations to the user."
In the same field of endeavor, (Moon) teaches:
"responsive to receiving an indication that a user accesses the online system, identifying one or more relevant nodes in the knowledge graph database for the user;" (Moon) teaches this limitation. (Moon) receives, from a client system associated with a user, a query from the user, accesses a knowledge graph of nodes and edges wherein each node corresponds to an entity, and determines, based on the query, one or more initial entities associated with the query, the initial entities being the relevant nodes and receipt of the user query being the indication of user access ((Moon), Abstract; claim 1, "receiving, from a client system associated with a user, a query from the user" and "determining, based on the query, one or more initial entities associated with the query").
"traversing the knowledge graph database to identify one or more item nodes connected to the relevant nodes; and" (Moon) teaches this limitation. (Moon) determines, by a conversational reasoning model, a path connecting the nodes corresponding to the initial entities to nodes corresponding to candidate entities based on one or more entity paths and one or more relation paths, thereby traversing the knowledge graph to identify candidate (item) nodes connected to the relevant nodes ((Moon), Abstract; claim 1, "determining, by a conversational reasoning model, a path connecting nodes corresponding to the one or more initial entities and nodes corresponding to candidate entities based on one or more entity paths and one or more relation paths").
"presenting items represented by the one or more item nodes as recommendations to the user." (Moon) teaches this limitation. (Moon) generates a response based on the initial entities and the candidate entities and sends instructions for presenting the response, representing the identified candidate nodes, to the user's client system ((Moon), Abstract, "generating a response based on the initial entities and the candidate entities" and "sending instructions for presenting the response to the client [system]").
(Yu) and (Moon) are analogous to the claimed invention as both are from the same field of endeavor of constructing and querying knowledge graphs of entities and relationships to serve item information to users. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine the large-language-model-constructed e-commerce intention knowledge graph of (Yu) with the query-driven knowledge-graph path traversal and response-presentation mechanism of (Moon), such that, responsive to a user accessing the online system, the system of (Yu) identifies relevant nodes, traverses the knowledge graph to identify connected item nodes, and presents the represented items to the user. The motivation to combine (Yu) and (Moon) is to enable the constructed knowledge graph of (Yu) to be used for retrieving and surfacing connected items in response to a user, since (Yu) already contemplates using the knowledge graph for downstream item recommendation ((Yu), page 8, § Recommendation) and (Moon) provides a known technique of traversing knowledge-graph entity and relation paths to identify candidate entities relevant to a user query and to present a corresponding response to the user ((Moon), Abstract; claim 1), yielding the predictable result of recommending items connected to the user's relevant nodes within the knowledge graph.
Claim 2 depends from claim 1. All limitations of claim 1 recited in claim 2 are rejected under the same rationale set forth for claim 1 above, over (Yu) in view of (Moon).
Regarding the further limitation added by claim 2, (Yu) teaches:
"wherein extracting the set of attributes comprises extracting one or more of: a characteristic of the item, a health-related concept of the item, a brand of the item, or a taxonomical category of the item." (Yu) teaches this limitation. The recited limitation is expressed in the alternative ("one or more of"); teaching any one of the enumerated attribute types satisfies the limitation under the broadest reasonable interpretation. (Yu) teaches extracting a characteristic of the item: (Yu) aligns the prompt templates with the ConceptNet HasProperty/HasA relations and extracts item characteristics such as "waterproof" as property assertions connected to the item ((Yu), page 4, § 3.3 and Table 2 "Prompts for different commonsense relations"; Figure 1, page 2, depicting a HasProperty assertion). (Yu) further teaches extracting a taxonomical category of the item: (Yu) aligns the prompts with the ConceptNet IsA relation and performs conceptualization to map the extracted entities to more high-level concepts, thereby extracting the item's taxonomical category ((Yu), page 4, § 3.3, listing IsA among the eighteen ConceptNet relations; § 3.1/§ 3.4, describing conceptualization to more abstract concepts; Figure 1, page 2, depicting an IsA assertion).
Claim 8 depends from claim 1. All limitations of claim 1 recited in claim 8 are rejected under the same rationale set forth for claim 1 above, over (Yu) in view of (Moon).
Regarding the further limitations added by claim 8, (Yu) teaches:
"generating a second prompt for input to the machine-learned language model, the second prompt specifying another request to infer a connection between a first entity represented by a first node and a second entity represented by a second node in the knowledge graph database;" (Yu) teaches this sub-limitation. (Yu) generates a further prompt for input to the large language model of the form "A user bought item 1 and item 2 because [GEN]," and aligns the prompt with the ConceptNet relations, the prompt requesting the model to infer a relation (connection) between a first item entity (item 1) and a second item entity (item 2) ((Yu), page 2, § 1, the co-buy prompt requesting the reason connecting the two items; page 4, § 3.3 and Table 2, aligning the prompts with eighteen ConceptNet relations to generate the assertion connecting the item pair).
"parsing the response from the model serving system to identify whether the connection exists between the first node and the second node; and" (Yu) teaches this sub-limitation. (Yu) structures the generated response and populates a plausibility score for the generated assertion, keeping only the assertions whose predicted plausibility score exceeds a threshold, thereby identifying whether the inferred connection between the two items is plausible, i.e., whether the connection exists ((Yu), page 4, § 3.4; page 5, § 3.4.2, populating the plausibility inference and retaining assertions whose predicted plausibility scores are above 0.5).
"responsive to the identification, storing the connection between the first node and the second node in the graph knowledge database." (Yu) teaches this sub-limitation. Responsive to the plausibility determination, (Yu) keeps only the plausible assertions in the final knowledge graph, thereby storing the connection between the two item nodes in the knowledge graph database ((Yu), page 5, § 3.4.2, only plausible assertions being kept in the final knowledge graph).
Claim 10 recites a non-transitory computer readable storage medium comprising stored program code instructions that, when executed, cause a processing system to perform operations corresponding to the method of claim 1. Claim 10 is rejected under the same rationale set forth for claim 1.
Claim 11 depends from claim 10 and recites the attribute-type alternatives of claim 2 in CRM form. Claim 11 is rejected under the same rationale set forth for claim 2.
Claim 17 depends from claim 10 and recites the second-prompt inter-node connection inference limitations of claim 8 in CRM form. Claim 17 is rejected under the same rationale set forth for claim 8.
Claim 19 is a system claim and is rejected under the same rationale set forth for claim 1.
Claim 20 is a computer system claim depending from computer system claim 19 (treated as such per the Claim Objection section above) and is rejected under the same rationale set forth for claims 2 and 11.
Claims 3, 4, 5, 6, 9, 12, 13, 14, 15, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Yu et al. (Yu), Non-Patent Literature, "FolkScope: Intention Knowledge Graph Construction for E-commerce Commonsense Discovery," arXiv:2211.08316v1, published November 15, 2022, in view of Moon et al. (Moon), U.S. Patent No. 11,442,992 B1, cited in the IDS filed on 2/21/25, and further in view of Cao et al. (Cao), U.S. Patent Application Publication No. 2020/0356598 A1.
Claim 3 depends from claim 1. All limitations of claim 1 recited in claim 3 are rejected under the same rationale set forth for claim 1 above, over (Yu) in view of (Moon).
Regarding the further limitations added by claim 3, (Yu) in view of (Moon) teaches:
"performing a training process to generate embeddings for the item node and the set of attribute nodes; and" (Yu) teaches this sub-limitation. (Yu) performs a training process that learns item embeddings from the constructed knowledge graph and represents the tail (attribute) nodes with learned node embeddings, thereby generating embeddings for the item node and the set of attribute nodes ((Yu), page 8, § Recommendation, learning item embeddings from the matched knowledge graph and representing tail-node embeddings using SentenceBERT representations).
"responsive to receiving a second indication that a second user accesses the online system," (Moon) teaches this sub-limitation. (Moon) receives, from a client system associated with a user, a query from the user, which is an indication that a user accesses the online system; the recited "second" user is a further instance of this same user-access operation for a subsequent user ((Moon), Abstract; claim 1, "receiving, from a client system associated with a user, a query from the user").
"identifying a relevant node in the knowledge graph database for the second user," (Moon) teaches this sub-limitation. (Moon) determines, based on the query, one or more initial entities associated with the query, the initial entity being the relevant node identified for the user ((Moon), Abstract; claim 1, "determining, based on the query, one or more initial entities associated with the query").
"presenting items represented by the subset of nodes as recommendations to the second user." (Yu) teaches this sub-limitation. (Yu) uses the learned node embeddings to train a recommendation model that recommends items to users, thereby presenting items represented by the nodes as recommendations to the user ((Yu), page 8, § Recommendation, using the item embeddings as features to train the recommendation model).
The combination of (Yu) and (Moon), however, does not teach:
"identifying a subset of nodes having embeddings less than a threshold distance from an embedding of the relevant node, and"
In the same field of endeavor, (Cao) teaches this sub-limitation. (Cao) obtains a dynamic embedding vector for each node of the relationship graph and determines similarity measures between the embedding vectors of two nodes, and when the similarity measure between the embedding vectors is greater than a predetermined threshold, identifies and acts on that node ((Cao), ¶ [0003], stating that when similarity measures between embedding vectors of two nodes are relatively high, products of one node can be recommended to the other node; and (Cao), FIG. 2 and associated description, "[d]etermine similarity measures between the fourth node and the fifth node based on a dynamic embedding vector of each node" and "[w]hen the similarity measures are greater than a predetermined threshold, push an [item]"). A similarity measure between embedding vectors that is greater than a predetermined threshold corresponds to an embedding distance that is less than a threshold distance; (Cao) thereby teaches identifying a subset of nodes whose embeddings are within a threshold distance of the embedding of a given node.
(Yu), (Moon), and (Cao) are analogous to the claimed invention as all three are from the same field of endeavor of generating node embeddings for a graph of entities and relationships and using those embeddings to identify and serve related items to a user. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine the large-language-model-constructed e-commerce intention knowledge graph and item-embedding recommendation of (Yu) and the query-driven relevant-node identification of (Moon) with the embedding-similarity-threshold node selection of (Cao), such that, responsive to a second user accessing the online system, the system identifies a relevant node for the second user and identifies a subset of nodes whose embeddings are within a threshold distance of the relevant node's embedding, and presents the items represented by that subset as recommendations. The motivation to combine (Yu), (Moon), and (Cao) is as recited by (Cao) (¶ [0003]), namely that recommending products to a node based on high embedding-vector similarity between nodes provides a known and predictable technique for selecting relevant items, such that one of ordinary skill would apply (Cao)'s embedding-distance thresholding to the node embeddings of (Yu) to select and recommend the most relevant connected items to the user, yielding the predictable result of embedding-based item recommendation within the knowledge graph.
Claim 4 depends from claim 3. All limitations of claim 3 recited in claim 4 are rejected under the same rationale set forth for claim 3 above, over (Yu) in view of (Moon) and further in view of (Cao).
Regarding the further limitation added by claim 4, (Yu) teaches:
"wherein an embedding for a node incorporates semantic information for text describing the node." (Yu) teaches this limitation. (Yu) represents the node embeddings of the tail nodes of the knowledge graph using SentenceBERT representations, which encode the semantic meaning of the text describing each such node; the node embedding therefore incorporates semantic information for the text describing the node ((Yu), page 8, § Recommendation, computing the node embeddings of the tail nodes from their SentenceBERT representations).
Regarding claim 5, the combination of (Yu) and (Moon) teaches learning node embeddings and recommending items generally ((Yu), page 8, § Recommendation). However, the combination does not teach the limitations set forth below, which are taught by (Cao).
In the same field of endeavor, (Cao) teaches:
"identifying a subset of nodes having embeddings less than a threshold distance from an embedding of the relevant node, and" (Cao) obtains a dynamic embedding vector for each node and determines similarity measures between the embedding vectors of two nodes, and when the similarity measure is greater than a predetermined threshold, identifies and acts on that node; a similarity measure greater than a threshold corresponds to an embedding distance less than a threshold distance ((Cao), ¶ [0003]; FIG. 2 and associated description).
"for one or more iterations:" (Cao) trains the embedding model through multiple times of optimization, i.e., over one or more iterations ((Cao), ¶ [0042]).
"for a selected node, applying parameters to embeddings for the selected node and one or more neighbor nodes connected to the selected node to generate a set of intermediate embeddings;" (Cao) inputs the input embedding vectors of a first (selected) node and its N neighboring nodes into the embedding model and applies first, second, and third weight matrices (parameters) to each input vector through a self-attention function, generating a set of transformed output vectors ((Cao), ¶¶ [0011]–[0012]; ¶ [Abstract]).
"combining the set of intermediate embeddings to generate an updated embedding for the node;" In (Cao), each function output vector is a weighted combination of the N+1 input vectors, and the model outputs the dynamic embedding vector of the node, the dynamic embedding being the updated embedding ((Cao), ¶ [0011]; ¶ [Abstract]).
"computing a loss function based on the updated embedding; and" (Cao) obtains a prediction result for each node from the dynamic (updated) embedding vector output by the embedding model, and the training unit performs model optimization based on the prediction result and a target for the node ((Cao), ¶ [0042]).
"backpropagating error terms from the loss function to update the parameters for a current iteration." (Cao) performs the model optimization based on the prediction result and target and returns optimized parameters to the embedding model, adjusting the parameters of the embedding model through the multiple times of optimization ((Cao), ¶ [0042]; ¶ [0014]).
(Yu), (Moon), and (Cao) are analogous to the claimed invention as all three are from the same field of endeavor of generating node embeddings for a graph of entities and relationships by aggregating information from neighboring nodes and using those embeddings to identify and serve related items to a user. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine the large-language-model-constructed knowledge graph and item recommendation of (Yu) and the query-driven relevant-node identification and presentation of (Moon) with the iterative neighbor-aggregation embedding training and embedding-similarity-threshold node selection of (Cao). The motivation to combine (Yu), (Moon), and (Cao) is as recited by (Cao) (¶ [0003]), namely that recommending products to a node based on high embedding-vector similarity provides a known and predictable technique for selecting relevant items, and that iteratively optimizing the embedding-model parameters produces node embeddings whose distances reflect node relationships ((Cao), ¶ [0042]), such that one of ordinary skill would apply (Cao)'s embedding training and embedding-distance thresholding to the node embeddings of (Yu) and (Moon) to select and recommend the most relevant connected items, yielding the predictable result of trained, embedding-based item recommendation within the knowledge graph.
Claim 6 depends from claim 1. All limitations of claim 1 recited in claim 6 are rejected under the same rationale set forth for claim 1 above, over (Yu) in view of (Moon).
The combination of (Yu) and (Moon) does not teach the limitation set forth below, which is taught by (Cao).
In the same field of endeavor, (Cao) teaches:
"storing the knowledge graph database as an array of lists," (Cao) represents the relationship graph as an adjacency list, which is an array of lists ((Cao), ¶ [0004], computing the embedding vector of each node based on an adjacency list of the relationship graph).
"wherein an element in the array represents a node and is associated with a list including neighbor nodes connected to the node." In (Cao)'s adjacency-list representation, each element corresponds to a node of the relationship graph and is associated with that node's neighboring nodes; (Cao) further determines, for each node, the N neighboring nodes connected to the node, i.e., the list of neighbor nodes associated with the element ((Cao), ¶ [0004], adjacency list of the relationship graph; ¶¶ [0006]–[0007], determining N neighboring nodes of a node, the neighboring nodes being nodes within a predetermined degree of the node).
(Yu), (Moon), and (Cao) are analogous to the claimed invention as all three are from the same field of endeavor of constructing and operating on a graph of entities and relationships to serve item information to users. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to store the knowledge graph database of (Yu) and (Moon) as an array of lists in which each element represents a node associated with a list of the node's neighbor nodes, as taught by (Cao). The motivation to combine (Yu), (Moon), and (Cao) is that representing the graph as an adjacency list is a known and efficient graph-storage structure that (Cao) relies upon to determine each node's neighboring nodes for computing node embeddings ((Cao), ¶ [0004] and ¶¶ [0006]–[0007]), such that one of ordinary skill would store the knowledge graph of (Yu) and (Moon) in this known structure to enable efficient retrieval of the neighbor nodes connected to a given node, yielding the predictable result of efficient neighbor lookup during graph traversal and embedding generation.
Claim 9 depends from claim 3. All limitations of claim 3 recited in claim 9 are rejected under the same rationale set forth for claim 3 above, over (Yu) in view of (Moon) and further in view of (Cao).
The combination of (Yu) and (Moon) teaches learning node embeddings from the knowledge graph and recommending items to a user ((Yu), page 8, § Recommendation). However, the combination of (Yu) and (Moon) does not teach the limitation set forth below, which is taught by (Cao).
In the same field of endeavor, (Cao) teaches:
"performing a retraining process to update the embeddings for the item node and the set of attribute nodes." (Cao) performs the embedding-training process through multiple times of optimization, adjusting the parameters of the embedding model and updating the input embedding vectors of the nodes at each optimization, thereby performing a training process that is repeated to update the node embeddings ((Cao), ¶ [0042], adjusting parameters through multiple times of optimization and updating the input embedding vector of each node each time of optimization; ¶ [0014], optimizing parameters of the embedding model and optimizing the input embedding vectors).
(Yu), (Moon), and (Cao) are analogous to the claimed invention as all three are from the same field of endeavor of generating node embeddings for a graph of entities and relationships and using those embeddings to identify and serve related items to a user. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine the knowledge-graph construction and item recommendation of (Yu) and the query-driven relevant-node identification and presentation of (Moon) with the iterative, repeated embedding-optimization process of (Cao), such that the node embeddings of (Yu) and (Moon) are updated through a repeated training process. The motivation to combine (Yu), (Moon), and (Cao) is that (Cao) teaches iteratively optimizing the embedding-model parameters and the node embeddings through multiple times of optimization ((Cao), ¶ [0042]), such that one of ordinary skill would apply (Cao)'s repeated embedding-optimization process to the node embeddings of (Yu) and (Moon) to keep the embeddings current as the underlying graph data changes, yielding the predictable result of updated node embeddings.
The combination of (Yu), (Moon), and (Cao) teaches a retraining process that updates the item- and attribute-node embeddings ((Cao), ¶¶ [0014], [0042]). However, the combination of (Yu), (Moon), and (Cao) does not teach the limitation set forth below.
"receiving feedback from the user on the presented items; and" Receiving feedback from a user on presented recommendation items—for example, in the form of clicks, selections, purchases, add-to-cart events, or ratings on the presented items—was notoriously well-known in the art of recommender systems before the effective filing date of the claimed invention, and using such user feedback as a training signal to retrain a recommendation model was likewise well-known and conventional. It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to receive feedback from the user on the items presented by the combination of (Yu), (Moon), and (Cao), such that one of ordinary skill would integrate the received user feedback on the presented items into the repeated embedding-optimization process of (Cao) to update the item- and attribute-node embeddings, in order to continuously improve the relevance and accuracy of the item recommendations by incorporating the user's observed responses to the presented items into the trained embeddings, yielding the predictable result of a recommender whose embeddings are refined over time based on user feedback.
Claim 12 depends from claim 10 and recites the training-and-second-user embedding-threshold recommendation limitations of claim 3 in CRM form. Claim 12 is rejected under the same rationale set forth for claim 3.
Claim 13 depends from claim 12 and recites the semantic-information node-embedding limitation of claim 4 in CRM form. Claim 13 is rejected under the same rationale set forth for claim 4.
Claim 14 depends from claim 12 and recites the iterative parameter-application, intermediate-embedding combination, loss-function, and backpropagation limitations of claim 5 in CRM form. Claim 14 is rejected under the same rationale set forth for claim 5.
Claim 15 depends from claim 10 and recites the array-of-lists storage limitation of claim 6 in CRM form. Claim 15 is rejected under the same rationale set forth for claim 6.
Claim 18 depends from claim 12 and recites the user-feedback and retraining-to-update-embeddings limitations of claim 9 in CRM form. Claim 18 is rejected under the same rationale set forth for claim 9.
Claims 7 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Yu et al. (Yu), Non-Patent Literature, "FolkScope: Intention Knowledge Graph Construction for E-commerce Commonsense Discovery," arXiv:2211.08316v1, published November 15, 2022, in view of Moon et al. (Moon), U.S. Patent No. 11,442,992 B1, cited in the IDS filed on 2/21/25, and further in view of Dong et al. (Dong), Non-Patent Literature, "AutoKnow: Self-Driving Knowledge Collection for Products of Thousands of Types," arXiv:2006.13473, published June 24, 2020.
Claim 7 depends from claim 1. All limitations of claim 1 recited in claim 7 are rejected under the same rationale set forth for claim 1 above, over (Yu) in view of (Moon).
The combination of (Yu) and (Moon) teaches obtaining descriptions for an item ((Yu), page 4, § 3.2 and § 3.3, obtaining item titles from the Amazon corpus). However, the combination does not teach the limitation set forth below, which is taught by (Dong).
In the same field of endeavor, (Dong) teaches:
"wherein obtaining the descriptions for the item further comprises obtaining the descriptions from a label of the item or instructions for the item provided by a retailer of the item." The limitation is recited in the alternative ("a label of the item or instructions for the item provided by a retailer of the item"); teaching either alternative satisfies the limitation under the broadest reasonable interpretation. (Dong) teaches obtaining the item descriptions from information for the item provided by a retailer of the item: (Dong) mines item knowledge from product profiles, such as titles and descriptions, contained in catalogs from e-Business retailers and contributed by the retailers, wherein retailers list the product features in the titles and descriptions ((Dong), page 2, § 1 (Introduction), stating that in catalogs from e-Business websites such as Amazon, eBay, and Walmart the data is often contributed by retailers, that retailers mainly list product features in titles and descriptions, and that the knowledge needs to be mined from the textual product profiles, e.g., titles and descriptions). The retailer-provided product descriptions of (Dong) thereby read on obtaining the descriptions from information for the item provided by a retailer of the item.
(Yu), (Moon), and (Dong) are analogous to the claimed invention as all three are from the same field of endeavor of collecting item information and constructing a knowledge graph of items and attributes for an e-commerce system. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to obtain the item descriptions used by (Yu) and (Moon) from the product descriptions provided by a retailer of the item, as taught by (Dong). The motivation to combine (Yu), (Moon), and (Dong) is that (Dong) teaches that retailers list product features in the titles and descriptions of their catalogs and that item knowledge is mined from these retailer-provided product profiles ((Dong), page 2, § 1), such that one of ordinary skill would obtain the item descriptions of (Yu) and (Moon) from the retailer-provided product descriptions to supply accurate, source-of-record item information for attribute extraction, yielding the predictable result of populating the knowledge graph from retailer-provided item descriptions.
Claim 16 depends from claim 10 and recites the retailer-provided item-description source limitation of claim 7 in CRM form. Claim 16 is rejected under the same rationale set forth for claim 7.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HUNG VAN LE whose telephone number is (571)270-0164. The examiner can normally be reached 8 a.m. - 5 p.m..
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, Cesar Paula can be reached at (571) 272-4128. 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.
/HUNG VAN LE/Examiner, Art Unit 2145
/CESAR B PAULA/Supervisory Patent Examiner, Art Unit 2145
Prosecution Insights