Detecting communities in social networks by local affinity propagation with grey relational analysis

Abstract

Purpose – The purpose of this paper is to discover social communities from the social networks by propagating affinity messages among members in a localized way. The affinity between any two members is computed by grey relational analysis method. Design/methodology/approach – First, the responsibility messages and the availability messages are restricted to be broadcasted only among a node and its neighbours, i.e. the nodes that connected to it directly. In this way, both the time complexity and the space complexity can be reduced to be near linear to the network size. The near-linear time and space complexity is quite important for social network analysis because social networks are generally very large. Second, instead of the widely used Euclidean distance, the grey relational degree is adopted in the calculation of node similarity, because the latter is more suitable for the discovery of the hidden relations among the nodes. On the basis of the two improvements, a new social community detection algorithm is proposed. Finally, experiments are conducted to verify the performance of the new algorithm. Findings – The new algorithm is evaluated by the experiments on both the real-world and the artificial data sets. The experimental results prove the proposed algorithm to be quite effective and efficient at community discovery. Practical implications – The algorithm proposed in the paper can be applied to discover communities in many social networks. After the recognition of the social communities, the authors can send advertisements, spot valuable customers or locate criminals more precisely. Originality/value – The new algorithm revises the affinity propagation progress to be localized to improve both time and space complexity. Furthermore, the grey relational analysis is applied to solve the complex relations among members of the social networks.