US 11,941,057 B2
Deep representation machine learned model for heterogeneous information networks
Zhanglong Liu, Fremont, CA (US); Ankan Saha, San Francisco, CA (US); Yiou Xiao, Sunnyvale, CA (US); Kathryn L. Evans, San Francisco, CA (US); Aastha Jain, Sunnyvale, CA (US); and Aastha Nigam, Sunnyvale, CA (US)
Assigned to Microsoft Technology Licensing, LLC, Redmond, WA (US)
Filed by Microsoft Technology Licensing, LLC, Redmond, WA (US)
Filed on Jun. 1, 2022, as Appl. No. 17/830,067.
Prior Publication US 2023/0394084 A1, Dec. 7, 2023
Int. Cl. G06F 16/901 (2019.01); G06F 16/28 (2019.01); G06N 3/044 (2023.01); G06N 3/08 (2023.01)
CPC G06F 16/9024 (2019.01) [G06F 16/288 (2019.01); G06N 3/044 (2023.01); G06N 3/08 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A system comprising:
a memory; and
a non-transitory computer-readable medium having instructions stored thereon, which, when executed by a processor, cause the system to perform operations comprising:
accessing data pertaining to a plurality of entities in an online network, the data including information about connections among the entities, the data including information about entities of multiple entity types;
for each entity to which the data pertains:
generating an entity graph indicating the connections between the entity and other entities in the online network;
for each entity in the data that the entity is connected with in the entity graph:
creating a modulo embedding for the entity;
creating a modulo embedding for the entity that the entity is connected with; and
concatenating the modulo embedding for the entity with the modulo embedding for the entity the entity is connected with;
selecting some but not all of the entities as representatives;
creating an anchor embedding using the representatives;
creating a meta-path embedding for each of a plurality of groupings of entities with direct connections with each other or indirect connections between entities of a same entity type up to two hops away from each other in the connections in the entity graph; and
passing the entity graph, the anchor embeddings, and the meta-path embeddings to a deep neural network to train the deep neural network to predict future connections among the entities, the passing including:
passing the entity graph, the anchor embeddings for entities of a first entity type, and meta-path embeddings created from the anchor embeddings of the first entity type to a first tower of the deep neural network; and
passing the entity graph, the anchor embeddings for entities of a second entity type, and meta-path embeddings created from the anchor embeddings of the second entity type to a second tower of the deep neural network.