Research On Node Classification Based On Graph Convolutional Network

With the recent advancement in the research of Graph Convolutional Network(GCN),a lot of attention has been paid to the problem of node classification as an important task in the data-mining network.The application and research of node classification is a very hot topic in recent years,and its application scenarios are extensive,including user classification in the social network,literature recognition in the citation network,mining protein interaction mechanism in the biochemistry field to study the causes of disease factors,etc.With the help of Neural Networks,the classification of nodes in topological graph data can effectively combine the node feature information and spatial structure information,and has achieved good performance.The research work carried out in this dissertation mainly focuses on a semi-supervised node classification method based on Graph Convolutional Network using the concept of spectral filtering and adjacency matrix.After constructing the graph structure data,the traditional Convolutional Neural Network is generalized based on spectrum theory,so that the irregular graph structure data can be directly processed.The simplified Graph Convolution based on spectral filtering uses the basic idea to capture local information and update the characteristics of central nodes with neighboring nodes.After the evaluation of results on three benchmark datasets(Citeseer,Cora,and Pub Med),the proposed method shows good performance against state-of-the-art algorithms.