US 12,475,376 B2
Generating node embeddings for multiple roles
Ryan A. Rossi, San Jose, CA (US); Iftikhar Ahamath Burhanuddin, Bangalore (IN); Gautam Choudhary, Sri Ganganagar (IN); Fan Du, Milpitas, CA (US); and Eunyee Koh, San Jose, CA (US)
Assigned to Adobe Inc., San Jose, CA (US)
Filed by Adobe Inc., San Jose, CA (US)
Filed on Jun. 22, 2022, as Appl. No. 17/846,160.
Prior Publication US 2023/0419115 A1, Dec. 28, 2023
Int. Cl. G06N 3/082 (2023.01)
CPC G06N 3/082 (2013.01) 20 Claims
OG exemplary drawing
 
1. A method comprising:
clustering, by a processing device, nodes of a graph into clusters;
computing, by the processing device, an initial role membership vector for each of the nodes based on the clusters;
generating, by the processing device, a first set of role embeddings for a first node of the nodes based on the initial role membership vector for the first node and nodes connected to the first node in the graph;
generating, by the processing device, a second set of role embeddings for a second node of the nodes based on the initial role membership vector for the second node and nodes connected to the second mode in the graph, in which, a number of role embeddings in the first set of role embeddings is different from a number of role embeddings in the second set of role embeddings;
determining, by the processing device, a node classification as a label for the first node based on the first set of role embeddings and the second set of role embeddings for the second node of the nodes; and
presenting, by the processing device, the label for the first mode for display in a user interface.
 
8. A system comprising:
a memory component; and
a processing device coupled to the memory component, the processing device to perform operations comprising:
clustering nodes of a graph into clusters;
computing an initial role membership vector for each of the nodes based on the clusters;
determining an updated role membership vector for each of the nodes by aggregating information from nodes connected to each of the nodes in the graph using an order invariant aggregator function;
generating a first set of role embeddings for a first node of the nodes by using the updated role membership vector for the first node to condition role embeddings of nodes connected to the first node in the graph;
determining an indication of a graph prediction based on the first set of role embeddings and a second set of role embeddings for a second node of the nodes; and
resending the indication for display in a user interface.