A Stable and Distributed Community Detection Algorithm Based on Maximal Cliques

Feng Gui,Yunlong Ma,Feng Zhang,Min Liu,Rong Yin,Weiming Shen

DOI: https://doi.org/10.1109/smc.2015.239

2015-01-01

Abstract:In the research area of community detection which aims at detecting some highly cohesive vertex subsets in social network, there mainly exist some problems, such as the algorithms with comparatively excellent quality of the final partitioning usually have high time complexity and some other fast algorithms often result in low quality of partitioning or other disadvantages. Nowadays, the increasing demands for community detection in large-scale social networks necessitate the use of distributed and scalable methods to detect communities in an effective and efficient manner. Label propagation algorithm (LPA), whose time complexity is O(m) on a network with m edges, is a near linear time algorithm to detect community effectively. Besides, owing to having good scalability, the parallel version of LPA (DLPA) is suitable for community detection in large-scale social networks. However, DLPA synchronously updates the vertices labels, which usually brings about label oscillations and results in low quality of partitioning. In this paper, we analyze the drawbacks of DLPA and propose a novel method C-DLPA, which combines DLPA with the notion of maximal cliques and at the same time utilizes a new updating mechanism that updating each node' label by probability of its adjacent nodes, to make final partitioning become more accurate and to avoid oscillations effectively. The experimental results show that C-DLPA has better performance is not only low time cost by as much to avoid oscillations but its community detection accuracy compared with DLPA.