A method for classifying papers with a graph convolutional network incorporating a capsule mechanism

Abstract

The invention discloses a method for classifying papers by a graph convolutional network integrated with a capsule mechanism. The method comprises the following steps: S1, constructing paper citation network data; S2, constructing a graph convolutional network model, inputting an adjacency matrix A and a node feature matrix X of a training set to train the graph convolutional network model, and obtaining a trained graph convolutional network model Model; S3. According to the model trained in S2, classify the papers and output the classification result. The invention integrates the capsule mechanism, regards the word vector of any paper as a capsule (Capsule), uses the dynamic routing algorithm to learn the similarity of word vectors between different papers, and makes the edge weight between the nodes corresponding to any two papers The size of is proportional to the similarity of the word vectors of the two papers. In the improved model, the weights of the edges on the graph network are changed. Compared with the model before the improvement, the accuracy is higher, the robustness is stronger, and it is less prone to overfitting.

Description

technical field [0001] The invention relates to the field of computer technology, in particular to a method for classifying papers by a graph convolutional network fused with a capsule mechanism. Background technique [0002] In the paper classification task of the existing Graph Convolution Network (GCN for short), the papers are used as nodes, the text information of the papers is used as the feature vector of the nodes, and there are edges between papers if there is a citation relationship. During model training, the weights of the edges between the nodes corresponding to the papers are fixed, and only the text information of the papers and the citation relationship between the papers are used to weight the feature vectors of the current node and the feature vectors of neighboring nodes. The result is used as the feature vector of the current node. Since papers of different categories may cite each other, for example, papers in the computer field may cite papers in the b...

AI Technical Summary

Problems solved by technology

Node 1 contains the characteristic information of the biological field, and it is difficult to accurately identify the category to which node 1 belongs during the model prediction process

Therefore, the method of summing and averaging is not reasonable

Images