SAG Cluster: An unsupervised graph clustering based on collaborative similarity for community detection in complex networks

Abstract

Many real-world social networks such as brain graph, protein structure, food web, transportation system, World Wide Web, online social networks exist in the form of a complex network. In such complex networks, pattern identification or community detection requires extra effort in which identifying community is a significant problem in various research areas. Most of the clustering methods on graphs predominantly emphasize on the topological structure without considering connectivity between vertices and not bearing in mind the vertex properties/attributes or similarity-based on indirectly connected vertices. A novel clustering algorithm SAG-Cluster with K-medoids framework presented for detecting communities using a collaborative similarity measure which considers attribute importance in case the pair of disconnected nodes. A novel path strategy using classic Basel problem for the indirectly connected node as well as balanced attribute similarity and distance function is proposed. On two real data sets, experimental results show the effectiveness of SAG-Cluster with the comparison of other relevant methods.