Community Detection Using Girvan–Newman and Kernighan–Lin Bipartition Algorithms

Abstract

Graph partitioning is a common problem in Graph Theory. Social Network Analysis becomes a very popular application of Web Structure Mining generally used for Community Detection (CD). CD is a graph partitioning of social network. The social networks are the series of people connected by a collection of social relations that are partitioned by a specific community. Communities are unstructured clusters of data in which subgraphs of nodes are tightly interlinked but broadly associated with each other. The features found from the network topology can determine community groups based on mutual interest, goals, specifications, professions, connectivity, relationships, and their hierarchical of the organization. The detection of communities by using these features can be beneficial in several applications primarily in social network which is the main motivation of proposed work. Some of the applications of social network analysis are used to keep track of communications between the user interactions, to find the collaborating structure of friend's within social media, to analysis the behavior of the user. The main objective of CD is to find clusters of nodes that are closely connected to each other and weakly connected to the rest of the graph. This work is proposed to study the effectiveness of Girvan--Newman and Kernighan--Lin bipartition algorithms on lesmis and email-Eu-core-department-labels network for detecting communities. The correctness of CD in networks with node features are evaluated by using NMI score measures on lesmis and email-Eu-core-department-labels network. Experimental results show that the Girvan--Newman algorithm provides best results for community detection on lesmis network with respect to corresponding ground-truth communities.