摘要
In the study of graph convolutional networks,the information aggregation of nodes is important for downstream tasks.However,current graph convolutional networks do not differentiate the importance of different neighboring nodes from the perspective of network topology when ag-gregating messages from neighboring nodes.Therefore,based on network topology,this paper proposes a weighted graph convolutional network based on network node degree and efficiency(W-GCN)model for semi-supervised node classification.To distinguish the importance of nodes,this paper uses the degree and the efficiency of nodes in the network to construct the impor-tance matrix of nodes,rather than the adjacency matrix,which usually is a normalized symmetry Laplacian matrix in graph convolutional network.So that weights of neighbor nodes can be as-signed respectively in the process of graph convolution operation.The proposed method is ex-amined through several real benchmark datasets(Cora,CiteSeer and PubMed)in the experimen-tal part.And compared with the graph convolutional network method.The experimental results show that the W-GCN model proposed in this paper is better than the graph convolutional net-work model in prediction accuracy and achieves better results.
基金
mainly supported by Fundamental Research Program of Shanxi Province(No.202203021211305)
Shanxi Scholarship Council of China(2023-013).
