AN NMF-Based Community Detection Method Regularized with Local and Global Information

Community identification is an important task in complex network analysis. Recently, Nonnegative Matrix Factorization (NMF) has been successfully used as an effective tool to discover community structures due to its powerful interpretability property. The main issues of this model are requiring a prior knowledge about the community structure and weak stability of the solution. To tackle these issues, in this paper a novel NMF-based method is proposed by incorporating both local and global information to the community identification process. The proposed method first, is mapped the graph into a data space using linear sparse coding. Then a novel metric is proposed to identify community centers that are used to form micro-communities. These micro-communities are further used to identify positive and negative edges that considered in global information regularization term. The local information of the graph is also used in the objective function our NMF-based model. Finally, the objective function is solved to form final communities. The experimental results on three real-world networks denote the superiority of proposed method compared to several well-known and state-of-the-art methods.