A graph-enhanced attention model for community detection in multiplex networks

Community detection is to divide a graph or a network into several components with high intra-edge density in a single-layer network. However, a multiplex network, consisting of multiple layers which share the same nodes but each owning different types of edges, brings new challenges for community detection. Although many classic algorithms and several multiplex embedding-based models try to address this problem, i.e. community detection in multiplex networks, they mostly focus on the local topological structures but ignoring the global information in each network layer, which is important for encoding intra-layer structural attributes and merging inter-layer semantic relations. In this paper, we propose a graph-enhanced attention model (GEAM) for community detection in multiplex networks by utilizing the above global information. In particular, the GEAM first constructs a layer contrastive learning module to encode node and graph embedding from the local and global graph view in each network layer. We also propose a self-attention adaptive fusion mechanism to learn a comprehensive version of node representation by fusing multiple layers. Finally, we propose an edge density-driven community detection module to train our GEAM in an end-to-end manner and output community divisions with strong modular structures. Experiments on both synthetic and real-world datasets validate the superiority of our GEAM over the state-of-the-art algorithms in terms of better community detection performance.