Measuring node dissimilarity and seeking communities in directed networks via signal copying

We consider the problem of seeking communities in directed networks. The common approach to this problem has been to simply ignore directions of edges and use algorithms developed for community detection in undirected networks, but this approach discards potentially indispensable information involved in the directions of edges. Here, we introduce an index for node dissimilarity in directed networks in the perspective of dynamics, which are driven by a novel parameter-free signal copying model, to incorporate information contained in the directions of edges. Our dissimilarity index gives a natural fusion between the local and global viewpoints of topological structure of directed networks. Thereby, we apply our index to uncover community structure in directed networks and conduct extensive experiments on synthetic and real-world ones. The results demonstrate that our method outperforms state-of-the-art methods.