Date: October 04, 2018 16:15–17:00
Place: Konemiehentie 2, Room T4 (Otaniemi)
Speaker: Cigdem Aslay
Title: Maximizing the diversity of exposure in a social network
Abstract: In this talk I will present our recent result that provides a novel approach to contribute towards bursting filter bubbles. We formulate the problem as a task of recommending news articles to selected users with the aim to maximize the overall diversity of information exposure in a social network. We consider a realistic setting where we take into account the political leanings of users and articles, and the probability of users to further share articles. We show that this problem is a challenging generalization of the influence maximization problem, which is NP-hard, and it corresponds to the problem of maximizing a monotone submodular function subject to a matroid constraint on the allocation of articles to users. We introduce the notion of random reverse co-exposure sets and a set of estimation techniques based on martingales for efficiently estimating expected diversity of exposure. Accordingly, we devise a scalable instantiation of the greedy algorithm that provides (1/2-epsilon)-approximation to the optimum with high probability.