Node Importance based Label Propagation Algorithm for overlapping community detection in networks

The enormous growth of the Web led to the birth of different network structures. Therefore, one of the important issues in the field of complex network analysis is to find and exploit the structure. Many studies have been carried out in this sense. The Label Propagation Algorithm (LPA) is among the most recognized approaches to detect disjointed communities. It is a simple and fast method, but its major disadvantage lies in its instability due to a random update. In this paper, we introduce a Node Importance based Label Propagation Algorithm (NI-LPA), a new algorithm for detecting overlapping communities in networks. As indicated in its name, NI-LPA is an improved version of LPA which maintains its simplicity and enhances its accuracy. In fact, we adopt the LPA strategy to allow a node to contain a set of labels. Moreover, the algorithm simulates a special propagation and filtering process using information deduced from the properties of nodes. Experimental results on artificial and real-world networks with different sizes, complexities and densities show the efficiency of our approach to detect overlapping communities.