Centrality-Aware Link Recommendations

Abstract

Link recommendations are critical for both improving the utility and expediting the growth of social networks. Most previous approaches focus on suggesting links that are highly likely to be adopted. In this paper, we add a different perspective to the problem by aiming at recommending links that also improve specific properties of the network. In particular, our goal is to recommend to users links that if adopted would improve their centrality in the network. Specifically, we introduce the centrality-aware link recommendation problem as the problem of recommending to a user u, k links from a pool of recommended links so as to maximize the expected decrease of the sum of the shortest path distances of $u$ to all other nodes in the network. We show that the problem is NP-hard, but our optimization function is monotone and sub-modular which guarantees a constant approximation ratio for the greedy algorithm. We present a fast algorithm for computing the expected decrease caused by a set of recommendations which we use as a building block in our algorithms. We provide experimental results that evaluate the performance of our algorithms with respect to both the accuracy of the prediction and the improvement in the centrality of the nodes, and we study the tradeoff between the two.