Community detection in networks based on information bottleneck clustering

Community detection is very crucial in social network research, and there is already a large body of work that investigates approaches to detecting communities, where clustering methods play an important role. In this paper, a partitional Information Bottleneck clustering based community Detection method (pIBD) is presented. The pIBD transforms a network graph from a unipartite network into a bipartite network, after which a matrix about the nodes is obtained. Based on the matrix, pIBD predicts k value and implement partition clustering under the information-theoretic framework. The k-value prediction defines external information loss and internal information loss, and estimates the number of clusters by calculating the crosspoint. Partitional clustering procedure starts from an initial random partition of network nodes, and implements an iterative process to reassign each node to optimal cluster. In order to test effectiveness of pIBD method, three real network datasets are selected. Experimental results show that our pIBD approach can achieve higher precision than aIBD method. 1553-9105/Copyright © 2015 Binary Information Press