Design of an overlapping community detection approach for social networks

Abstract

A social network is a representation of individuals and organizations that depicts their connections and interactions. Social network analysis (SNA) is a multidisciplinary field of research that reveals the patterns in the network and how these participating individuals interact. Detecting communities over a network helps in understanding network structure which further enhances its applicability in various domains ranging from physical to life sciences. One of the key challenges of social networks analysis is detection of communities involved in information diffusion. This thesis proposed an effective overlapping community detection method SLPA-IF1. Initially nodes label initialization is done during pre-processing of data. Label updation and propagation is performed during the evolution phase which consists of selection of listener node, speaker and listener rule. Speaker rule is modified to consider the mean of occurring frequency of labels instead of random label selection. Additionally, we have also proposed a new measure named label specificity for listener rule. The proposed method leads to more accurate label selection during detection of communities over a network. The proposed SLPA-IF1 method has outperformed other state-of- the-art methods selected for overlapping community detection. The core idea of changing the evolution phase by proposing new measures for speaker rule results in more accurate performance of the proposed method while retaining linear computational time. F1 score and linear run time quantifies the better performance of the proposed SLPA-IF1 method for the detection of communities over a network. newline