Detecting Communities from Signed Network Based on Local Search

Xueyan Liu,Bo Yang,Xuehua Zhao,Yi Yang

DOI: https://doi.org/10.2991/icence-16.2016.103

2016-01-01

Abstract:Many researchers have begun to study signed networks which are widely existed in real world. In the signed network, the links are labeled the positive or negative sign to represent the active or passive relation between individuals, such as trusted or distrusted relation in social networks. Communities mining is still a great challenge to the domain of signed networks because of negative links. Unlike communities of unsigned networks, positive links mainly occur in the communities and negative links tend to occur between the communities in the signed networks. Nowadays, many methods which are based on global search for signed network community have been raised, and most of these methods require the global information at each iteration. Besides, determining the number of communities is an important problem for current algorithm for the lack of priori knowledge. To address above problems, a novel community detection method based on local information, is proposed for signed networks in this paper. The proposed method mainly includes two steps. In the first step, the number of communities is determined in terms of the centrality of nodes. In the second step, the local objective function is optimized by the local information of nodes, so the global objective function can also be optimized indirectly. Finally, the communities in signed networks are efficiently found. To validate the proposed method, the comparisons are made with other methods in the synthetic and real signed networks. The experimental results indicate that communities in signed networks can be efficiently found by the proposed method.