Adaptive Graph Learning for Semi-supervised Classification of GCNs

Graph convolutional networks (GCNs) have achieved great success in social networks and other aspects. However, existing GCN methods generally require a wealth of domain knowledge to obtain the data graph, which cannot guarantee that the graph is suitable. In this paper, we propose adaptive graph learning for semi-supervised classification of GCNs. Firstly, the hypergraph is used to establish the initial neighborhood relationship between data. Then hypergraph, sparse learning and adaptive graph are integrated into a framework. Finally, the suitable graph is obtained, which is inputted into GCN for semi-supervised learning. The experimental results of multi-type datasets show that our method is superior to other comparison algorithms in classification tasks.