Reveal Community Structure by Local Similarity and Hierarchical Clustering

Guijie Zhang,Yu Xin,Shuai Wang,Suxia Chen,Xu Yu

doi:10.1109/access.2019.2940213

Abstract

Community structure is one of the most important topological properties of complex networks, which can help us to understand the functions and guide the development of networks. In this article, a community detection algorithm is proposed based on local similarity and hierarchical clustering. Local similarity is used to measure link similarities instead of node similarities in order to form a similarity metric. Hierarchical clustering is used to gather all the links to form a hierarchical tree, and then cut the tree with the optimization value of modularity to get the community structure. Experiments on real-world and generated benchmark networks show the significant performance of the algorithm both in accuracy and efficiency.

Highlights

Community detection plays an important role in understanding the structure of social network functions, uncovering hidden patterns and predicting the behavior of the network
Definition 5 (Link Community): Unlike traditional community detection algorithm which takes nodes as research object, we focus on links to detect the community
A similarity metric is proposed based on the local link structure to establish the similarity matrix of links

Summary

INTRODUCTION

Community detection plays an important role in understanding the structure of social network functions, uncovering hidden patterns and predicting the behavior of the network. The time consuming of our algorithm constitutes mainly of four parts, mapping the node-node adjacency matrix to the link-link adjacency matrix, calculating the local similarity of the link graph, getting the clustering tree by the hierarchical clustering and transforming the link community to the node community and its optimization. Because both of the real networks and BA scale-free networks have the similar long tail phenomenon, the growth of the number of edges performs a linear relationship with the growth of the number of nodes and the growth rate of links is about ten times to that of nodes. Because most networks in the real world are sparse ones, we can regard the complexity as O(n2) according to the relationship between nodes and links in the sparse network

NUMERICAL EXPERIMENTS

VALIDATION METRICS

COMMUNITY DETECTION RESULTS ANALYSIS ON LFR BENCHMARK

CONCLUSION