Overlapping Community Extraction: A Link Hypergraph Partitioning Based Method

Real-world networks often contain communities with pervasive overlaps such that nodes simultaneously belong to several groups. Community extraction, emerging in recent years, is considered to be a promising solution for finding meaningful communities from social networks. In this paper, we explore overlapping community extraction from a link partitioning perspective. First, we define the local link structure composed of a set of closely interrelated links, by extending the similarity of link-pairs to that of a group of links. Second, based upon our prior work, we transform the problem of mining local link structures into a pattern mining problem, and thus present an efficient mining algorithm. Third, we propose to use the hypergraph to assemble all local link structures, and employ hMETIS for hypergraph partitioning. Finally, based on extracted link communities, we restore the membership of nodes in the original graph owing to its links. Experimental results on various real-life social networks validate the effectiveness of the proposed method.