Finding overlapping communities based on information fusion in social network

Abstract

Social network is becoming indispensable of people's lives in recent years. Community detection on real network continues to be a hotspot in data ming domain. As users may join multiple social circles and interest communities, and an abundance of information can be a reflection of users' preference, heterogeneous information fusion and overlapping community detection are two key issues researchers have to face. This paper presents an overlapping community detection model IF-COPRA that incorporated heterogeneous information into an integrated user adjacency diagram, based on which multi-label propagation and overlapping community detection are fulfilled. Before label propagation, user adjacency diagram is pruned to eliminate noisy relationship and vertices are processed in degree otder to enlarge the influence of high degree vertices and improve the robustness of IF-COPRA. Experiments on real world data sets demonstrate that IF-COPRA model performances better than baseline algorithms in most cases.