Detecting Communities in Signed Networks based on Structural Balance Theory and Graph Convolution Network

Singed networks describe the positive and negative interactions between components in real complex systems, contain rich information and are helpful to comprehensively and deeply mine connection patterns between individuals. It is a challenging task to detect the community structure accurately and quickly in a signed network. In this paper, we propose a community detection algorithm based on convex non-negative matrix factorization and graph convolutional neural network to extract communities from signed networks, which combines the aggregation process of graph convolutional networks with structural balance theory to make better use of the positive and negative connections, and designs a loss function combining the relaxed modularity and graph regularization to train the model in an end-to-end fashion. Finally, the effectiveness of the proposed algorithm is verified by experiments on artificial networks and real networks.