Semi-supervised community detection based on discrete potential theory

In recent studies of the complex network, most of the community detection methods only consider the network topological structure without background information. This leads to a relatively low accuracy. In this paper, a novel semi-supervised community detection algorithm is proposed based on the discrete potential theory. It effectively incorporates individual labels, the labels of corresponding communities, to guide the community detection process for achieving better accuracy. Specifically, a number of vertices with user-defined labels are first identified to act as unit elementary charges which can generate different electrostatic fields. Then, community detection can be translated into a potential transmission problem. By formulating the problem using combinational Dirichlet, labels of those unlabeled vertices can be determined by the labels for which the greatest potential is calculated. Finally, a better community partition can be obtained. Our extensive numerical experiments in both artificial and real networks lead to two key observations: first, individual labels play an important role in community detection; and second, our proposed semi-supervised community detection algorithm outperforms existing counterparts in both accuracy and time complexity, especially for obscure networks.