A Signal-Strategy-Based Spectral Clustering Method for Community Detection in Complex Networks

The community detection has been one of the core subjects in complex networks. Spectral clustering is an efficient method widely used in this field. In spectral clustering, the Laplacian matrix should be built with similarity matrix, however, similarity matrix is often been replaced by adjacency matrix because few appropriate ways could be used to measure the node similarity in a complex network. As the solution, an appropriate measure of similarity should be proposed to build Laplacian matrix. Signal strategy has been proved to be an efficient method reflecting the relationships between nodes in complex network, and the relationship could be considered as a reasonable scale. This paper presents a semi-supervised spectral approach for community detection, the proposed method uses signal strategy to generate the Laplacian matrix, and utilizes prior knowledge to further guarantee the detection performance. Experiments results showed that the proposed method gave excellent performance on real world network and Lancichinetti-Fortunato-Radicchi (LFR) benchmark, with comparison of other spectral and non-spectral community detection methods.