Proximity-based group formation game model for community detection in social network

Abstract

As an increasing amount of network data emerges, especially for online daily social networks, a prominent question arises is how to observe the formed community structure. In this paper, we propose a proximity-based group formation game model, called PBCD, to detect communities in social networks. PBCD’s motivation is based on an empirical observation that the higher number of shared communities gives rise to the higher second-order pairwise proximity. By illustrating the generation process of second-order pairwise proximity, PBCD achieves a convincible performance on community detection. Furthermore, we formulate a two-step non-cooperative game model to illustrate the evolution process of community structure in each period. Using a well-designed potential function, we provide a strict proof that the subgame of first step is resembled with a classic potential game. Finally, we discuss an extended version by introducing community interaction probability matrix into PBCD to deal with the community detection task. Comprehensive experiments conducted on real-world social networks show that our approach can achieve good performance both in terms of detection accuracy and execution efficiency.