An energy model for network community structure detection

Abstract

Community detection problem has been studied for years, but no common definition of community has been agreed upon till now. Former modularity based methods may lose the information among communities, and blockmodel based methods arbitrarily assume the connection probability inside a community is the same. In order to solve these problems, we present an energy model for community detection, which considers the information of the whole network. It does the community detection without knowing the type of network structure in advance. The energy model defines positive energy produced by attraction between two vertices, and negative energy produced by the attraction from other vertices which weakens the attraction between the two vertices. Energy between two vertices is the sum of their positive energy and negative energy. Computing the energy of each community, we may find the community structure when maximizing the sum of these communities energy. Finally, we apply the model to find community structure in real-world networks and artificial networks. The results show that the energy model is applicable to both unipartite networks and bipartite networks, and is able to find community structure successfully without knowing the network structure type. © Springer-Verlag 2013.