Abstract: Community detection is one of the key research directions in the study of complex networks. Most of the existing work focuses on network topology but ignores the dynamic process on the network. Thus, a dynamic diffusion community detection algorithm based on central node is proposed. First, a node centrality measure metric is proposed based on the number of non-backtracking path. Then, in order to model the multi-scale social interaction mode occurring on the network, a new edge membership vector is designed to represent the community belonging of nodes which links the central node with community detection. Besides, a dynamic system is designed to represent the dynamic distribution process of community members to complete overlapping community detection. Finally, the proposed algorithm is applied to real networks and artificial networks to verify its effectiveness. The experimental results show the proposed algorithm has great advantages in detection accuracy.

Key words: complex networks, community detection, overlapping structures, non-backtracking matrix, membership vector