A Graph Distance Based Structural Clustering Approach for Networks

In the era of information explosion, structured data emerge on a large scale. As a description of structured data, network has drawn attention of researchers in many subjects. Network clustering, as an essential part of this study area, focuses on detecting hidden sub-group using structural features of networks. Much previous research covers measuring network structure and discovering clusters. In this paper, a novel structural metric "Graph Distance" and an effective clustering algorithm GRACE are proposed. The graph distance integrates local density of clusters with global structural properties to reflect the actual network structure. The algorithm GRACE generalizes hierarchical and locality clustering methods and outperforms some existing methods. An empirical evaluation demonstrates the performance of our approach on both synthetic data and real world networks.