US 11,487,791 B2Granted

Latent feature extraction from a network graph

AssigneeMicrosoft Technology Licensing, LLC

InventorsMatthew Hsing Hung Walker, Myunghwan Kim, Yiou Xiao, Yafei Wang, Florent Bekerman

Life of this patent

Estimating the term…

Loading patent…

Abstract

Techniques for generating latent representations for entities based on a network graph are provided. In one technique, an artificial neural network is trained based on feature vectors of entities and feature vectors of neighbors of those entities. The neighbors are determined based on a graph of nodes representing the entities. Two nodes are connected if, for example, the corresponding entities are connected in an online network, one entity transmitted an online communication to the other entity, or one entity interacted with content associated with the other entity. Once trained, the artificial neural network is used to generate latent representations for entities represented by the graph. Latent representations may be used in multiple ways. For example, a similarity between two latent representations may be used to determine an order of candidate content items to present to an entity corresponding to one of the latent representations.

Claims20

Dependent claims

A method comprising:

for a first particular node of a first plurality of nodes in a graph of affinities between at least one of people and organizations:
identifying a first plurality of neighbor nodes that are within N connections of the first particular node in the graph, where N is a positive integer less than or equal to 3;
identifying a first plurality of feature vectors that represent attribute values of a first particular neighbor node of the first plurality of neighbor nodes;
identifying a first feature vector of the first particular node;
generating a first aggregated vector for the first particular node by aggregating the first feature vector and the first plurality of feature vectors;
generating a training instance that comprises the first aggregated vector, a second aggregated vector generated for a second particular node of the first plurality of nodes that is different than the first particular node, and a label indicating whether the first particular node and the second particular node are neighbors in the graph;
adding the training instance to a set of training data;
training an artificial neural network based on the set of training data including the training instance;
for a second particular node of a second plurality of nodes in the graph:
identifying a second plurality of neighbor nodes of the second particular node;
identifying a second plurality of feature vectors, each corresponding to a different neighbor node of the second plurality of neighbor nodes;
identifying a second feature vector of the second particular node;
generating a third aggregated vector based on the second feature vector and the second plurality of feature vectors;
using the trained artificial neural network, generating an embedding based on the third aggregated vector and the training instance;
storing the embedding in association with the second particular node;
wherein the method is performed by one or more computing devices.

13 more elements

One or more non-transitory storage media storing instructions which, when executed by one or more processors, cause:

for a first particular node of a first plurality of nodes in a graph of affinities between at least one of people and organizations:
identifying a first plurality of neighbor nodes that are within N connections of the first particular node in the graph, where N is a positive integer less than or equal to 3;
identifying a first plurality of feature vectors that represent attribute values of a first particular neighbor node of the first plurality of neighbor nodes;
identifying a first feature vector of the first particular node;
generating a first aggregated vector for the first particular node by aggregating the first feature vector and the first plurality of feature vectors;
generating a training instance that comprises the first aggregated vector, a second aggregated vector generated for a second particular node of the first plurality of nodes that is different than the first particular node, and a label indicating whether the first particular node and the second particular node are neighbors in the graph;
adding the training instance to a set of training data;
training an artificial neural network based on the set of training data including the training instance;
for a second particular node of a second plurality of nodes in the graph:
identifying a second plurality of neighbor nodes of the second particular node;
identifying a second plurality of feature vectors, each corresponding to a different neighbor node of the second plurality of neighbor nodes;
identifying a second feature vector of the second particular node;
generating a third aggregated vector based on the second feature vector and the second plurality of feature vectors;
using the trained artificial neural network generating an embedding based on the third aggregated vector and the training instance;
and
storing the embedding in association with the second particular node.

13 more elements

Dependencies are inferred from the claim text.