	US 12,217,157 B2
	System, method, and computer program product for dynamic node classification in temporal-based machine learning classification models
Jiarui Sun, Urbana, IL (US); Mengting Gu, Stanford, CA (US); Michael Yeh, Newark, CA (US); Liang Wang, San Jose, CA (US); and Wei Zhang, Fremont, CA (US)
Assigned to Visa International Service Association, San Francisco, CA (US)
Appl. No. 18/271,301
Filed by Visa International Service Association, San Francisco, CA (US)
PCT Filed Jan. 30, 2023, PCT No. PCT/US2023/011830 § 371(c)(1), (2) Date Jul. 7, 2023, PCT Pub. No. WO2023/147106, PCT Pub. Date Aug. 3, 2023.
Claims priority of provisional application 63/304,771, filed on Jan. 31, 2022.
Prior Publication US 2024/0078416 A1, Mar. 7, 2024
Int. Cl. G06N 3/04 (2023.01); G06F 17/16 (2006.01); G06N 3/049 (2023.01)

CPC G06N 3/049 (2013.01) [G06F 17/16 (2013.01)]

20 Claims

1. A computer-implemented method comprising:

receiving, with at least one processor, graph data of a discrete time dynamic graph comprising a plurality of graph snapshots, wherein each graph snapshot of the plurality of graph snapshots is associated with an instance of the discrete time dynamic graph at a time step in a first time period, and wherein each graph snapshot of the plurality of graph snapshots comprises a plurality of nodes, a plurality of edges, and a node attribute matrix;

receiving, with the at least one processor, a plurality of node classifications associated with all nodes in the discrete time dynamic graph, wherein each node classification of the plurality of node classifications is associated with a node of the plurality of nodes in each graph snapshot of the plurality of graph snapshots;

converting, with the at least one processor, the discrete time dynamic graph to a time-augmented spatio-temporal graph based on the plurality of edges of each graph snapshot of the plurality of graph snapshots;

generating, with the at least one processor, an adjacency matrix of at least one adjacency matrix based on a temporal walk of the time-augmented spatio-temporal graph, the adjacency matrix comprising a plurality of adjacency matrices positioned along a diagonal of the adjacency matrix, wherein each adjacency matrix of the plurality of adjacency matrices is associated with a graph snapshot of the plurality of graph snapshots at a separate time step in the first time period, and wherein the adjacency matrix of the at least one adjacency matrix comprises at least one copied matrix of one or more adjacency matrices of the plurality of adjacency matrices;

generating, with the at least one processor, at least one adaptive information transition matrix based on the at least one adjacency matrix;

determining, with the at least one processor, a plurality of feature vectors based on the plurality of nodes and the node attribute matrix of each graph snapshot of the plurality of graph snapshots, wherein each feature vector of the plurality of feature vectors is associated with a node of the discrete time dynamic graph;

generating, with the at least one processor, a plurality of initial node representations, wherein each initial node representation of the plurality of initial node representations is based on a feature vector of the plurality of feature vectors and a node associated with the feature vector;

propagating, with the at least one processor, the plurality of initial node representations across a plurality of information propagation layers using the at least one adaptive information transition matrix to produce a plurality of final node representations based on an output of a final layer of the plurality of information propagation layers; and

classifying, with the at least one processor, at least one node of the discrete time dynamic graph in a time step of a second time period subsequent to the first time period based on the plurality of final node representations.