Overlapping community detection in directed heterogeneous social network

Abstract

In social networks, users and artifacts (documents, discussions or videos) can be modelled as directed bi-type heterogeneous networks. Most existing works for community detection is either with undirected links or in homogeneous networks. In this paper, we propose an efficient algorithm OcdRank (Overlapping Community Detection and Ranking), which combines overlapping community detection and community-member ranking together in directed heterogeneous social network. The algorithm has low time complexity and supports incremental update. Experiments show that our method can detect better community structures as compared to other existing community detection methods.