Community Detection Based on Directed Weighted Signed Graph Convolutional Networks

Currently, community detection in signed networks has become a popular research topic due to the widespread use of signed networks for modeling relationships among entities in the real world. However, most of these existing graph neural network based methods still have two limitations. The first one is that these methods are not applicable for the directed and weighted signed networks, and the second one is that the GNN methods cannot consider the network community structure, resulting in the learned node features failing to capture community-oriented characteristics. In view of these limitations, this paper proposes a directed weighted signed graph convolutional network for community detection called DWSGCN. For the first limitation, we introduce novel aggregation strategies based on two social psychological theories and construct a weighted adjacency matrix to fully extract the direction and weight information of links. In order to obtain the community-oriented node embedding, a novel modularity maximization loss is designed for signed networks and combined with a structure loss to jointly optimize DWSGCN. Finally, we obtain the community results in an end-to end manner. Extensive experiments demonstrate the superiority of DWSGCN over most state-of-the-art approaches.