Feature Graph Augmented Network Representation for Community Detection

Community detection plays an important role in understanding complex networks. Many traditional embedding-based community detection methods only focus on the relations between nodes in the topology space (i.e., topology graph). Besides, there are also some works that consider the feature embedding of nodes to further improve the detection performance. However, most of them ignore the relationships between nodes in the feature space (i.e., feature graph). To address this issue, in this article, we construct the feature graph from the features of nodes to capture the relations between nodes in the feature space, and incorporate it with the topology graph and the feature embedding, leading to the novel feature graph augmented network representation for community detection (FGCD) method. Specifically, FGCD extracts the embeddings of topology graph, node features, and feature graph, respectively, and ensembles them by a layerwise fusion method with an attention mechanism. Extensive experiments on 11 real-world datasets show that FGCD outperforms most existing state-of-the-art algorithms, which well demonstrates its superiority.