A Model of Opinion Dynamics for Community Detection in Graphs

Abstract

In this paper, we propose a new approach to the problem of community detection in graphs. It is based on a model of opinion dynamics with decaying confidence. This model is a multi-agent system where each agent receives the opinions of its neighbors and then updates its opinion by taking a weighted average of its own opinion and those of its neighbors that are within some confidence range. The confidence ranges are getting smaller at each time step: an agent gives repetitively confidence only to the neighbors that approach sufficiently fast its own opinion. Under that constraint, global consensus may not be achieved and the agents may only reach local agreement organizing themselves in communities. A characterization of these communities is given in terms of eigenvalues of normalized Laplacian matrices of graphs. This shows that our model of opinion dynamics with decaying confidence provides an appealing approach to community detection in graphs. Experimental results show that our approach is also effective. Copyright © IFAC 2010