Graph node classification method and application

Abstract

The invention belongs to the technical field of artificial intelligence, and particularly relates to a graph node classification method and application. The problem that graph representations of all convolutional layers are not combined at the same time in an existing method is solved. The invention provides a graph node classification method, which comprises the following steps of: performing graph convolution operation in graph data to obtain graph representation; calculating a state represented by each graph; evaluating the smooth saturation of the graph representation state of each layer, and counting the number of convolution layers; aiming at all nodes of the whole graph, linearly combining all layers of graph representations with the smooth saturation to obtain new node representations; performing classification through linear transformation and a normalized exponential function according to the new node representation to obtain a classification result; performing back propagation according to an error between the classification result category label and a real category label; and repeating the previous steps until convergence, stopping iteration, and outputting a graph node classification result. And high-order information and low-order information are effectively combined, so that the accuracy of model classification prediction is improved.

Description

technical field [0001] The present application belongs to the field of artificial intelligence technology, and in particular relates to a graph node classification method and application. Background technique [0002] The ACT model is an adaptive computing time model that automatically adapts to different network depths according to the input information at different times and the current time step. However, their method is not suitable for tasks related to graph-structured data, and can only be used for tasks related to time series. Moreover, in the time series task, the vertical stacking of layers in the network structure is to learn more abstract features of the current time sequence elements, which is not equivalent to the neighbor node convolution of the nodes in the graph network data. The AGC model is a graph convolutional network that adaptively controls the convolutional layer based on node similarity, which is a variant of the graph convolutional network and machi...

AI Technical Summary

Problems solved by technology

Combining the representations of all convolutional layers, highlighting low-order information while taking into account high-order information, and the existing methods have not combined the graph representation of all convolutional layers at the same time. This application provides a graph node classification method and application

Images