OverCite: Finding overlapping communities in citation network

Citation analysis is a popular area of research, which has been usually used to rank the authors and the publication venues of research papers. With huge number of publications every year, it has become difficult for the users to find relevant publication materials. One simple solution to this problem is to detect communities from the citation network and recommend papers based on the common membership in communities. But, in today´s research scenario, many researchers´ fields of interest spread into multiple research directions resulting in an increasing number of interdisciplinary publications. Therefore, it is necessary to detect overlapping communities for relevant recommendation. In this paper, we represent publication information as a tripartite `Publication Hypergraph´ consisting of authors, papers and publication venues (conferences/journals) in three partitions. We then propose an algorithm called `OverCite´, which can detect overlapping communities of authors, papers and venues simultaneously using the publication hypergraph and the citation network information. We compare OverCite with two existing overlapping community detection algorithms, Clique Percolation Method (CPM) and iLCD, applied on citation network. The experiments on a large real-world citation dataset show that OverCite outperforms other two algorithms. We also present a simple paper search and recommendation system. Based on the relevance judgements of the users, we further prove the effectiveness of OverCite over other two algorithms.