Community Detection Methods On Social Networks

The subject of community detection on social networks has important practicalsignificance. Many data sets from real world can be easily converted to networks ina very natural way. Since the Internet entered the era of Web2.0, a variety of socialapplications arise, providing a wealth of research material for the study of onlinesocial networks. Analysis of such networks and the discovery of implicit patternsbehind them can be very valuable.Many studies have demonstrated the existence of the modular structure onsocial networks. This paper focuses on the problem of non-overlapping communitydetection on social networks, introduces the research background and surveys thelandscape of this issue, and also elaborates the basis of the community detectionproblem, discusses the evaluation criteria and the data sets, especially the LFRbenchmark graphs.Then, several classic community detection methods with great significance areintroduced and compared with each other. Experiments of scalability and efficiencyare conducted on both real networks and synthetic networks.On the basis of comparative analysis, this paper proposes an improved labelpropagation algorithm, Functional Attenuation Label Propagation Algorithm(FALPA), to avoid some of the problems of traditional algorithms, such as therandomness of the LPA. Experiments have shown that FALPA achieves best resultsthan other label propagation methods.Against the community detection problem of online microblogging socialnetworks, an algorithm called Multi-Information Community Detection Algorithm(MICDA) is proposed. This Algorithm can integrate topological, content andinteractive information to improve the results of community detection algorithms bysimplifying and reconstructing the original network. Experiments show that anetwork processed by MICDA can reveal the community structure better than theoriginal one.