Emergence of core-periphery structure from local node dominance in social networks

There has been growing evidence recently for the view that social networks can be divided into a well connected core, and a sparse periphery. This paper describes how such a global description can be obtained from local "dominance" relation ships between vertices, to naturally yield a distributed algorithm for such a decomposition. It is shown that the resulting core describes the global structure of the network, while also preserving shortest paths, and displaying "expander-like" properties. Moreover, the periphery obtained from this de composition consists of a large number of connected com ponents, which can be used to identify communities in the network. These are used for a `divide-and-conquer´ strategy for community detection, where the peripheral components are obtained as a pre-processing step to identify the small sets most likely to contain communities. The method is illustrated using a real world network (DBLP co-authorship network), with ground-truth communities.