Hybrid machine learning approach for community and overlapping community detection in social network

Abstract

AbstractSocial Network Analysis (SNA) is one of the significant fields of sociology, which probes many researchers toward it. In SNA, overall social network (such as Facebook, Twitter, and so on) structure is learned by community detection process. However, dynamic nature and involvement of numerous users in social network limits the accuracy of community detection process. In this research work, we propose a novel dual examining and hybrid machine learning approaches for community and overlapping community detection. We collect recent tweets from twitter social network in real time as training dataset. An undirected graph is constructed over collected twitter dataset to enhance the community detection process. Upon constructed graph, informative examining and structural examining processes are applied. Informative examining results with informative factor whereas structural examining identifies influencing node. Structural examining is performed using Particle Swarm Optimization algorithm based on centrality factor and node strength factor. By utilizing results from dual examining processes, community detection is performed by a hybrid machine learning approach Naïve Bayes with Firefly Algorithm. Overlapping community detection is followed up to the community detection using Fuzzy with neural network classifier. Here, Jaccard Similarity factor, overlapping coefficient, and modularity are considered in fuzzy with neural network. Extensive simulation results show better performance than single examining methods in terms of accuracy of mutual information, ratio of community, run time, and average degree in proposed hybrid machine learning approach‐based community detection.