Greedy Modularity Graph Clustering for Community Detection of Large Co-Authorship Network

Ahmed F Al-Mukhtar,Eman S Al-Shamery

doi:10.14419/ijet.v7i4.19.28058

Abstract

Social networks as a domain of complex networks that can be represented as graphs according to the patterns of connections among their elements. Social Communities are a set of nodes with denser connections inside community structures than outside. The goal of graph clustering is to divide the large graph into many clusters depending on multiple similarity criteria. In this work an improved version of the Louvain method is proposed, the Greedy Modularity Graph Clustering for Community Detection of Large Co-AuthorshipNetwork (GMGC)which introduces a new concept of weighted edges to enhance the accuracy of the Community Discovery for the large networks. The method is compared with other states of art methods mainly, Vertices Similarity First and Community Mean (VSFCM), and Generalized Louvain method for community detection in large networks (FKCD). Extensive experimental results have been madeon different datasets. The experimental results showed that the proposed method outperforms the other states of arts comparative methods according to the modularity optimization and community partitions evaluations measures.

Highlights

Graph plays crucial roles in many domains of complex networks such as bioinformatics, social networks, sensor networks,and web, that because such kinds of systems can be visualized as connected nodes based on a specific relationship[1], [2]
Different graph clustering approaches have been developed to solve the community detection problem including spectral clustering, hierarchical clustering, both models cannot scale for large graphs.The fast, simple approach called label propagation in [4], which can execute in linear time. suitable for large networks the method leads to merging many smaller communities into a big cluster
The results are obtained by analyzing several datasets using the GMGC method other stats of art methods mainly, Clause, Moore and Newman (CNM) based on modularity, Vertices Similarity First and Communities Mean (VSFCM) based on weighted Clustering coefficient connection Degree (WCCD) and agglomerative clustering, Louvain method (LM), and the Fast κpath Community Detection (FKCD) a generalized LM based

Summary

Introduction

Graph plays crucial roles in many domains of complex networks such as bioinformatics, social networks, sensor networks,and web, that because such kinds of systems can be visualized as connected nodes based on a specific relationship[1], [2]. Community detection as an essential domain in social networks aims to distribute vertices based on their locations in the modules, by determining the structural boundaries of each module[3], [4]. Different graph clustering approaches have been developed to solve the community detection problem including spectral clustering, hierarchical clustering, both models cannot scale for large graphs.The fast, simple approach called label propagation in [4], which can execute in linear time. Suitable for large networks the method leads to merging many smaller communities into a big cluster. A community detection method is proposed for large weighted networks. Called Greedy Modularity Graph Clustering for Community Detection of Large Co-Authorship Network (GMGC).

Methods

Results

Conclusion