Communities Mining and Recommendation for Large-Scale Mobile Social Networks

Two well-known phenomena are observed in social networks. One is the tendency of users to connect with similar users, leading to the emergence of communities. The other is that certain users belong to multiple communities simultaneously. Understanding these phenomena is the major concern of social network analysis. In this work we focus on overlapping communities detection and personalized recommendation methods. We propose an algorithm with the property which takes closeness and influence of users into account for community detection, and utilizes semantic analysis and statistical analysis for the personalized recommendation. Our contributions include adopting the idea of greedy expansion involved with Clique Theory, extending PageRank to detect communities, and creating recommender from the view of semantics and statistics. In experiments, the algorithm is verified in terms of F1-measure, AP and MAP. The results show that our proposed algorithm can outperform the state-of-the-art methods.