Research On Detecting Overlapping Communities In Complex Networks Based On Link Clustering

Networks have become a key approach to understan ding systems of interacting objects.One crucial step when studying the structure and dynamics of networks is to identify communities.Few of current methods can detect overlapping and hierarchical communities simultaneously.Link Communities reinvent comm unities as groups of links rather than nodes and show that this unorthodox approach successfully reveal overlapping and hierarchical organization in networks.But,link community method has the drawbacks of low communities accuracy and weak overlapping nod es detecting ability.So the research on improving the performance of link community method will be very valuable in both theory and application.Focus on the drawbacks of low communities accuracy and weak overlapping nodes detecting ability of link commun ity method,three improving strategies are proposed in this paper,including two new link similarity definitions Inte-K and Plus-K that take into account the influence of key stone node and an improved single-linkage hierarchical clustering method that “a witnessed node”(a common neighbor)condition attached.Several networks were generated by LFR benchmark by adjusting the parameters u and c.Experiments on those networks show that the improved approaches can get a better balance among several criteria contrast to traditional Link Community method and hold the advantage of simple to implement.Following the improved Link Community methods,this method is applied to community detecting in bipartite graphs.A link clustering method is proposed to detect communities in bipartite graph.First,the similarity of two edges in bipartite graph is defined including punishing the popularity of key-stone node and all nodes;then hierarchical clustering is performed based on the similarity.The dendrogram shows the overlapping and hierarchical organization in bipartite graphs.Results of experiments on synthetic and real networks show that the method proposed can detect overlapping communities of bipartite networks efficiently,with the benefit of finding hierarchical structure of communities.