A Community Detection Method Based on the Subspace Similarity of Nodes in Complex Networks

Many real-world networks have a topological structure characterized by cohesive groups of vertices. Community detection aims at identifying such groups and plays a critical role in network science. Till now, many community detection methods have been developed in the literature. Most of them require to know the number of communities and the low accuracy in the complex networks are the shortcomings of most of these methods. To tackle these issues in this paper, a novel community detection method called CDNSS is proposed. The proposed method is based on the nodes subspace similarity and includes two main phases; seeding and expansion. In the first phase, seeds are identified using the potential distribution in the local and global similarity space. To compute the similarity between each pair, a specific centrality measure by considering the sparse linear coding and self-expressiveness ability of nodes. Then, the nodes with best focal state are discovered which guarantees the stability of solutions. In the expansion phase, a greedy strategy is used to assign the unlabeled nods to the relevant focal regions. The results of the experiments performed on several real-world and synthetic networks confirm the superiority of the proposed method in comparison with well-known and state-of-the-art community detection methods.