GCN and Non-negative Matrix Factorization-Based Community Detection Mechanism

Non-negative matrix factorization (NMF) is widely used in the field of community discovery because of its effectiveness and easy interpretation. However, most of the existing NMF-based community discovery methods are linear and cannot effectively deal with the nonlinear characteristics of complex networks, resulting in community discovery performance that needs to be further improved. To solve this problem, a graph convolutional network (GCN)-enhanced nonlinear NMF community discovery method NMFGCN is proposed. NMFGCN contains two main modes: GCN and NMF, where GCN is used to learn the network node representation, and NMF takes the node representation as input to obtain the community representation of the network. In addition, a joint optimization method is proposed to train NMFGCN, which not only enables NMFGCN to have nonlinear feature representation ability, but also enables GCN and NMF to promote each other and obtain better community segmentation results. Extensive experiments on synthetic networks and real networks show that NMFGCN outperforms current NMF-based community discovery methods, thus proving that NMFGCN can indeed improve the performance of NMF community discovery methods. In addition, NMFGCN also outperforms DeepWalk and LINE commonly used graph representation learning methods.