Overlapping Community Detection in Directed Graphs with Considering Local Fitness

Abstract

While graphs and networks are ubiquitous in our daily life and are attracting widely concern in the research area, detection of communities is a nontrivial subject. More recently, the focus in this subject has been switched to the detection of overlapping communities. In this paper, we focus on overlapping community detection in the directed graphs. We propose a greedy algorithm to detect the overlapping communities. In that algorithm, the detection process is split into three stages: seed selection, community detection and community merging. In the seed selection stage, we select the most seeds with importance computation. And in the second stage, we use the local fitness function to expand the communities which are initialized by the seed nodes. And we use the third stage to decrease the too small communities. At last, we conduct the experiment on the data generated by LFM benchmark, the results show that our method performs well in the directed graphs, and it performs better over 5% then the compared algorithm. We also discuss the stability of the algorithm to check whether the performance is sensitive with the parameter.