Graph neural network training method and device

Abstract

The embodiment of this specification provides a training method of a graph neural network. The method includes: first obtaining a relational network diagram, including a plurality of object nodes corresponding to a plurality of business objects; then, for each object node, merging its node characteristics with those of its neighbor nodes to obtain the fusion characteristics of the object node , the multiple fusion features corresponding to the multiple object nodes form a fusion feature matrix; use the graph neural network to perform graph embedding processing on the relationship network graph to obtain multiple embedding vectors corresponding to the multiple object nodes, the graph neural network An activation function is included in the network, and multiple prediction results are determined based on the plurality of embedding vectors; and a product matrix of the fusion feature matrix before and after processing by the activation function is determined; based on the product matrix, multiple prediction results and business labels, The training gradient of the parameters in the graph neural network is determined, and then based on the training gradient, the parameters in the graph neural network are updated.

Description

technical field [0001] The embodiments of this specification relate to the field of computer technology, and in particular to a training method and device for a graph neural network. Background technique [0002] The relationship network diagram is a description of the relationship between entities in the real world, and is currently widely used in various computer information processing. Generally, a relational network graph includes a set of nodes and a set of edges. Nodes represent entities in the real world, and edges represent connections between entities in the real world. For example, in a social network, people are entities, and relationships or links between people are edges. [0003] In many cases, it is hoped to analyze the topological characteristics of nodes, edges, etc. in the relational network graph, and extract effective information from them. The computing method to realize this kind of process is called graph computing. Typically, it is desired to repres...

AI Technical Summary

Problems solved by technology

However, in the training phase of the graph neural network, due to the need to train millions of parameters and the highly non-convex nature of the objective function, conventional training consumes a large amount of computing resources, and it is difficult to guarantee the convergence of the training and the training output. Embedding Effects of Graph Neural Networks

Images