An Algorithm for Identifying Useful Structure in Graphs Clustering

A new clustering method for graphs which is based on Breadth-First-Search is proposed. The clusters are of various structures such as cliques, stars, including hubs and outliers. Hubs and Outliers should be marked because they play very different roles in the network. Traditional algorithms may detect some useful structures, but they tend to fail to find hubs and outliers. We propose a clustering algorithm to solve the problem. Both synthetic networks and real datasets experiments demonstrate that apart from finding hubs and outliers, the algorithm also outperforms other algorithms in speed with a running time O(n) on a network with n nodes and m links, which is much faster than the fastest modularity-based algorithm O(mdlogn) (where d is the depth of the dendrogram describing the hierarchical cluster structure).