Community detection in dynamic networks via adaptive label propagation.

Jihui Han,Wei Li,Weibing Deng,Zhu Su,Longfeng Zhao,Yijiang Zou,Jürgen Kurths

doi:10.1371/journal.pone.0188655

Abstract

An adaptive label propagation algorithm (ALPA) is proposed to detect and monitor communities in dynamic networks. Unlike the traditional methods by re-computing the whole community decomposition after each modification of the network, ALPA takes into account the information of historical communities and updates its solution according to the network modifications via a local label propagation process, which generally affects only a small portion of the network. This makes it respond to network changes at low computational cost. The effectiveness of ALPA has been tested on both synthetic and real-world networks, which shows that it can successfully identify and track dynamic communities. Moreover, ALPA could detect communities with high quality and accuracy compared to other methods. Therefore, being low-complexity and parameter-free, ALPA is a scalable and promising solution for some real-world applications of community detection in dynamic networks.

Highlights

IntroductionCommunity structure is a prominent feature of networks and has received much attention in recent years
Many real-world systems can be represented as networks [1,2,3,4], in which nodes represent individuals and edges represent the relationships or interactions between individuals, such as the Internet [5], friendship networks [6], collaboration networks [7], food webs [8, 9], and metabolic networks [10, 11].Community structure is a prominent feature of networks and has received much attention in recent years
The average number of involved nodes in each time step is 23.7, which is tiny compared to the network size, so our algorithm can efficiently respond to the changes in network topology

Summary

Introduction

Community structure is a prominent feature of networks and has received much attention in recent years. It deepens our understanding of the underlying structure of many real-world networks [5,6,7,8,9], and promises a variety of practical applications ranging from the determination of functional modules within neural networks to the analysis of communities on the Internet. Most of the methods treat the network as a static one which is derived from aggregating data during a long period of time

Methods

Results

Conclusion