An Efficient Method Based on Label Propagation for Overlapping Community Detection

In this paper we focus on a research of the overlapping community structures. Towards this goal, we propose DOCLPA (Detecting Overlapping Community based Label Propagation Algorithm), a fast label propagation method is designed to detect overlapping community in GraphChi (a parallel sliding windows method) with only network topology. Here the label initialization is designed to reduce labels redundancy by local maximal cliques. We focus in particular on propagation characteristics between nodes, as far as possible, let the label propagation process be similar to the real information dissemination. Crucially, the DOCLPA method can detect high quality communities adaptively and does not require limited parameter. The experimental findings on real-world networks show that DOCLPA is effectiveness and accuracy in comparison with GCE, COPRA and LPA. Furthermore, by adjusting the overlapping coefficient to reveal the behavior characteristics of overlapping nodes, we prove that overlapping communities on the real-world networks are very naturally social phenomenon. In addition, our experiments on social, product, collaboration and communication networks demonstrate that overlapping nodes often represent active users or products who involve multiple different communities.