# US 2023/0410131 A1 · Hierarchy aware graph representation learning > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2023/0410131 A1 - Document ID: US20230410131A1 - Status: Published application - Kind: A1 - Application number: 18/242,294 - Filed: 2023-09-05 - Published: 2023-12-21 - Earliest claimed: 2021-02-11 - Assignees: Yahoo Assets LLC - Applicants: Yahoo Assets LLC - Inventors: Suleyman Cetintas; Pengyang Wang - Web page: https://patentgrep.com/patent/US20230410131A1 ## CPC - G06Q 30/0202: Market predictions or forecasting for commercial activities - H04L 67/02: based on web technology, e.g. hypertext transfer protocol [HTTP] - H04L 67/1089: {Hierarchical topologies} - H04L 67/1097: for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS] - H04L 67/535: {Tracking the activity of the user (network monitoring arrangements H04L43/00; recording of computer activity G06F11/34)} - G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR - H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M) ## Abstract A method includes executing operations to generate a first enhancement function based on a parent-child link in a content hierarchy including a link between a parent node in a first level of the content hierarchy to a child node in a second level of the content hierarchy below the first level. A second enhancement function is generated based on a sibling link in the content hierarchy including a link between a sibling node in a third level of the content hierarchy and a sibling node in the third level of the content hierarchy sharing a common parent node with the first sibling node in a fourth level of the content hierarchy above the third level. A user content consumption metric is generated based on the first and second enhancement functions. A content list including a set of candidate content items ranked based on the user content consumption metric is generated. ## Claims ### Claim 1 A method for hierarchy-aware graph representation learning, comprising: executing, on a processor of a computing device, instructions that cause the computing device to perform operations, the operations comprising: generating a first enhancement function for graph representation learning based on a parent-child link in a content hierarchy, wherein the first enhancement function corresponds to a combination of at least a first calculation associated with parent nodes sampled from a first set and a second calculation associated with non-parent nodes sampled from a second set; generating a second enhancement function for graph representation learning based on a sibling-sibling link in the content hierarchy, wherein the second enhancement function corresponds to a combination of at least a third calculation associated with a similarity of first samples of a sibling set and a fourth calculation associated with a similarity of second samples of a non-sibling set; inputting at least one of user feature data or one or more content hierarchies to a machine learning classifier comprising a neural network; and using the machine learning classifier, comprising the neural network, to generate a user content consumption metric based on the first enhancement function corresponding to the parent-child link and the second enhancement function corresponding to the sibling-sibling link. ### Claim 2 (depends on 1) The method of claim 1, wherein generating the user content consumption metric comprises: generating point-of-interest metrics; and generating the user content consumption metric based on the point-of-interest metrics. ### Claim 3 (depends on 1) The method of claim 1, wherein generating the user content consumption metric comprises: generating click-through rate metrics; and generating the user content consumption metric based on the click-through rate metrics. ### Claim 4 (depends on 1) The method of claim 1, wherein generating the user content consumption metric comprises: generating a unified enhancement function based on the first enhancement function and the second enhancement function; and generating the user content consumption metric based on the unified enhancement function. ### Claim 5 (depends on 1) The method of claim 1, wherein generating the user content consumption metric comprises: generating a unified enhancement function based on the first enhancement function, the second enhancement function, and a third enhancement function based on hop distances between nodes in the content hierarchy; and generating the user content consumption metric based on the unified enhancement function. ### Claim 6 (depends on 1) The method of claim 1, wherein generating the user content consumption metric comprises: assigning a first weight to the first enhancement function to generate a first weighted enhancement function; assigning a second weight to the second enhancement function to generate a second weighted enhancement function; assigning a third weight to a third enhancement function to generate a third weighted enhancement function, wherein the third enhancement function is based on hop distances between nodes in the content hierarchy; generating a unified enhancement function based on the first weighted enhancement function, the second weighted enhancement function, and the third weighted enhancement function; and generating the user content consumption metric based on the unified enhancement function. ### Claim 7 (depends on 1) The method of claim 1, wherein generating the first enhancement function comprises: generating a loss function based on a first set of parent-child nodes and a second set of non-parent nodes; and generating the first enhancement function based on the loss function. ### Claim 8 (depends on 1) The method of claim 1, wherein generating the second enhancement function comprises: generating a loss function based on a first set of sibling-sibling nodes and a second set of non-sibling nodes; and generating the second enhancement function based on the loss function. ### Claim 9 A computing device comprising: a processor; and memory comprising processor-executable instructions that when executed by the processor cause performance of operations for hierarchy-aware graph representation learning, the operations comprising: enhancing mutual information metrics for graph representation learning based on a parent-child link in a content hierarchy to generate enhanced mutual information metrics, wherein the enhanced mutual information metrics correspond to a combination of at least a first calculation associated with parent nodes sampled from a first set and a second calculation associated with non-parent nodes sampled from a second set; enhancing similarity metrics for graph representation learning based on a sibling-sibling link in the content hierarchy to generate enhanced similarity metrics, wherein the enhanced similarity metrics correspond to a combination of at least a third calculation associated with a similarity of first samples of a sibling set and a fourth calculation associated with a similarity of second samples of a non-sibling set; enhancing hop metrics based on nodes in the content hierarchy to generate enhanced hop metrics; inputting at least one of user feature data or one or more content hierarchies to a machine learning classifier comprising a neural network; and using the machine learning classifier, comprising the neural network, to generate a user content consumption metric based on the enhanced mutual information metrics, the enhanced similarity metrics, and the enhanced hop metrics. ### Claim 10 (depends on 9) The computing device of claim 9, wherein: the user content consumption metric comprises a point-of-interest metric. ### Claim 11 (depends on 9) The computing device of claim 9, wherein: the user content consumption metric comprises a click-through rate metric. ### Claim 12 (depends on 9) The computing device of claim 9, wherein generating the user content consumption metric comprises: generating a unified enhancement function based on the enhanced mutual information metrics, the enhanced similarity metrics, and the enhanced hop metrics; and generating the user content consumption metric based on the unified enhancement function. ### Claim 13 (depends on 9) The computing device of claim 9, wherein generating the user content consumption metric comprises: assigning a first weight to the enhanced mutual information metrics to generate weighted enhanced mutual information metrics; assigning a second weight to the enhanced similarity metrics to generate weighted enhanced similarity metrics; assigning a third weight to the enhanced hop metrics to generate weighted enhanced hop metrics; generating a unified enhancement function based on the weighted enhanced mutual information metrics, the weighted enhanced similarity metrics, and the weighted enhanced hop metrics; and generating the user content consumption metric based on the unified enhancement function. ### Claim 14 (depends on 9) The computing device of claim 9, wherein the operations comprise: generating the mutual information metrics using a loss function based on a first set of parent-child nodes and a second set of non-parent nodes. ### Claim 15 (depends on 9) The computing device of claim 9, wherein the operations comprise: generating the mutual information metrics using a loss function based on a first set of sibling-sibling nodes and a second set of non-sibling nodes. ### Claim 16 A non-transitory machine readable medium having stored thereon processor-executable instructions that when executed cause performance of operations for hierarchy-aware graph representation learning, the operations comprising: generating a first enhancement function for graph representation learning based on a relationship between a first node in a first level of a content hierarchy and a second node in a second level of the content hierarchy below the first level, wherein the first enhancement function corresponds to a combination of at least a first calculation associated with parent nodes sampled from a first set and a second calculation associated with non-parent nodes sampled from a second set; generating a second enhancement function for graph representation learning based on hops between nodes in the content hierarchy, wherein the second enhancement function corresponds to a combination of at least a third calculation associated with a similarity of first samples of a sibling set and a fourth calculation associated with a similarity of second samples of a non-sibling set; and inputting at least one of user feature data or one or more content hierarchies to a machine learning classifier comprising a neural network. ### Claim 17 (depends on 16) The non-transitory machine readable medium of claim 16, comprising: using the machine learning classifier, comprising the neural network, to generate a user content consumption metric based on the first enhancement function and the second enhancement function. ### Claim 18 (depends on 17) The non-transitory machine readable medium of claim 17, wherein generating the user content consumption metric comprises: generating a unified enhancement function based on the first enhancement function and the second enhancement function; and generating the user content consumption metric based on the unified enhancement function. ### Claim 19 (depends on 17) The non-transitory machine readable medium of claim 17, wherein: the user content consumption metric comprises a point-of-interest metric. ### Claim 20 (depends on 17) The non-transitory machine readable medium of claim 17, wherein: the user content consumption metric comprises a click-through rate metric.