Abstract:
In today's connected world there are complex systems and the science of network has been given significant time and advances to understand those complex structures. The community structure of real networks and finding those communities become a motivation to study. One of the most important methodology in community detection is clustering, in which an independent part of a network can be found and the overlapping among those compartments of a network can be uncovered. These days the bipartite network has been given a separate attention and category to understand the structure within it. The network formulated by bipartite network data are presented in two different vertices and the interaction is between those vertices. There are many algorithms proposed to understand and uncover the community structure in the bipartite network. Most of the analysis are done using a small network data or synthetic data. Here we are trying to assess different algorithms which are used to detect communities in the bipartite network. Using conventional comparison methods, we evaluate the performance of each algorithm. The considered algorithms are BCFinder, BigClam, CODA and PaNDa+. The first two algorithms are clique based algorithms while BigClam uses cluster affiliation model, and CODA is an extension of BigClam which include directed networks arises from cohesive and 2-mode communities. PaNDA+ a Top-k pattern mining algorithm can be used to detect overlapping communities from bipartite networks. We test the algorithms against artificial benchmark networks and also with networks which have ground truth dataset.
