Estimating Top-k Betweenness Centrality Nodes in Online Social Networks

Abstract

Betweenness centrality is a widely used network measure in social network analysis. There are many algorithms for calculating or approximating this measure, but most of these algorithms assume that all network information is known. Therefore, since we can obtain little information for online social networks because of security and privacy concerns, we must consider a crawling-based algorithm. Herein, we propose a new crawling-based algorithm for estimating the top-k nodes with the highest betweenness centrality in online social networks. The proposed algorithm approximates the ego betweenness centrality of the nodes sampled via a random walk and approximates the top-k nodes with the highest betweenness centrality in a graph as the top-k nodes with the highest approximated ego betweenness centrality in the sampled nodes. This algorithm makes the same number of requests to an application programming interface as existing algorithms because we reuse sampled nodes for approximation. Our experimental results show that the proposed algorithm can estimate the top-k nodes of real social networks more accurately than existing methods when sample size is very small.