Parallel computation of hierarchical closeness centrality and applications

It has long been an area of interest to identify important vertices in social networks. Closeness centrality is one of the most popular measures of centrality of vertices. Generally speaking, it measures how a node is close to all other nodes on average. However, closeness centrality measures the centrality from a global view. Consequently, in real-world networks that is normally composed by some communities connected, using closeness centrality may suffer from the flaw that local central vertices within communities are neglected. To resolve this issue, we propose a new centrality measure, Hierarchical Closeness Centrality (HCC), to depict the local centrality of vertices. Experiments show that comparing with closeness centrality, HCC is a better index in finding most influential vertices and community detection. Furthermore, we present a parallel algorithm for HCC computation, by well analyzing the independence between vertices in the computation procedure. Extensive experiments on real-world datesets demonstrate that the parallel algorithm can greatly reduce the computation time compared to trivial algorithms.