Comparing community identification algorithms in online social networks

Abstract

Online social networks have been provided the new relationship formation to people`s lives. In online social networks, users can build relationships with people whom they cannot meet in the off-line world. They not only interact with each other, but also form communities. The measurement and analysis of online social networks are important to improve current systems of online social networks. Online social networks have been analyzed from a macroscopic and microscopic view. Especially, we can find how actively users interact with each other in communities and how the evolution of communities affects properties of online social networks from a microscopic view by analyzing communities. Several algorithms have been proposed to detect communities in networks. However, these algorithms using modularity as the measure identify different communities depending on the order of input data. In this paper, we compare three community identification algorithms: CNM, Wakita, and MAM algorithm. We evaluate scalability, variance in modularity, and consistency in clustering of three algorithms with randomly ordered datasets of various networks. To our best knowledge, this is the first work to compare community identification algorithms from a view of the consistency. We show that the modularity of three algorithms depends on the order of input data and three algorithms have the consistency problem.