Mining Overlapping Communities in Real-world Networks Based on Extended Modularity Gain

Abstract

Detecting communities plays a vital role in studying group level patterns of a social network and it can be helpful in developing several recommendation systems such as movie recommendation, book recommendation, friend recommendation and so on. Most of the community detection algorithms can detect disjoint communities only, but in the real time scenario, a node can be a member of more than one community at the same time, that leads to overlapping communities. A novel approach is proposed to detect such overlapping communities by extending the definition of newman’s modularity for overlapping communities. The proposed algorithm is tested on LFR benchmark networks with overlapping communities and on real-world networks. The performance of the algorithm is evaluated using popular metrics such as ONMI, Omega Index, F-score and Overlap modularity and the results are compared with its competent algorithms. It is observed that extended modularity gain can detect highly modular structures in complex networks with overlapping communities.