PSCAN: A Parallel Structural Clustering Algorithm for networks

Abstract

Network clustering is receiving increasing attention in many application areas. However, as the size of data increases, traditional algorithms fail to deal with the large-scale data. Using distributed computers to handle such a problem in a parallel manner is an ideal solution. Many parallel clustering algorithms have been proposed, however, few are proposed for networks with the clustering criteria that uses the structure of networks and connectivity of vertices. In this paper, we present a parallel network clustering algorithm based on a structural similarity measure which can identify not only clusters in networks but also hubs and outliers. Our Parallel Structural Clustering Algorithm adapts the Structural Clustering Algorithm for Networks (SCAN) to a parallel environment. We prove that the clustering result of PSCAN is consistent with that of SCAN. Complexity analysis and experiments show that the proposed algorithm can solve the network clustering problem in low time complexity on large-scale networks.