Influence Maximization Revisited

Yihan Geng, Kunyu Wang, Ziqi Liu, Michael Yu, Jeffrey Xu Yu

Abstract: Influence Maximization (IM) has been extensively studied, which is to select a set of k seed users from a social network to maximize the expected number of influenced users in the social network. There are many approaches proposed under a cascade model to find such a single set of k seed users. Such a set being computed may not be unique, as it is most likely that there exist more than one set, \(S_1, S_2, \cdots \), each of them leads to the same IM, given a social network exhibits rich symmetry as reported in the literature. In this paper, first, we study how to select a set of k seed users from a set of seed \(k'~(\ge k)\) users which can be either a union of sets of seed users, \(\mathbb {S} = \bigcup _i S_i\), where \(S_i\) is a set of k seed users, or simply a set of seed users of size \(k'~(\ge k)\) being computed, based on cooperative game using Shapley value. Second, we develope a visualization system to explore the process of influence spreading from topological perspective, as IM only gives the expected number of influenced users without much information on how influence spreads in a large social network. We conduct experimental studies to confirm the effectiveness of the seed users selected in our approach.