Learning Influential Relationships for Implicit Influence Maximization in Unknown Networks

Social networks have been a common medium for spreading information among users. Inspired by this phenomenon, business companies consider ‘hiring’ some influential users in the social network to expand product promotion or brand influence. This problem can be computationally modeled as the Influence Maximization (IM) problem, which aims to find the best set of influential users to activate under a limited budget. Most existing solutions to the IM problem assume that the topological structure of a target network and the influential probabilities among users are provided in advance. However, acquiring complete influential relationships among users in an online social network can be challenging in many real-world applications. In this paper, we model the IM problem in such unknown social networks as the Implicit Influence Maximization (IIM) problem. To solve the IIM, we propose an ensembling approach consisting of two major components, i.e., Edge Explorer and Influence Probability Estimator. The Edge Explorer explores influential edges among users, while the Influence Probability Estimator estimates the influence probabilities associated with edges. The experimental results illustrate the effectiveness of the proposed approach.