Dynamic graph convolutional networks by semi-supervised contrastive learning

Guolin Zhang, Zehui Hu, Guoqiu Wen, Junbo Ma, Xiaofeng Zhu

Abstract: Highlights•The proposed model can exploit the implicit and correlative feature information between the nodes.•A high-quality feature graph is dynamically constructed from the input node features and iteratively refined through a specially designed semi-supervised contrastive learning method.•Two co-attention modules are proposed to fuse the embeddings from the given topology and the learned feature graph to extract the most correlative information.•A series of experiments on seven benchmark datasets demonstrate the effectiveness of the proposed model.