Learning Automata Based Approach for Influence Maximization Problem on Social Networks

Abstract

Influence maximization problem aims at targeting a subset of entities in a network such that the influence cascade being maximized. It is proved to be a NP-hard problem, and many approximate solutions have been proposed. The state-ofart approach is known as CELF, who evaluates the marginal influence spread of each entity by Monte-Carlo simulation and picks the most influential entity in each round. However, as the cost of Monte-Carlo simulations is in proportion to the scale of network, which limits the application of CELF in real-world networks. Learning automata (LA) is a promising technique potential solution to many engineering problem. In this paper, we extend the confidence interval estimator based learning automata to S-model environment, based on this, an end-to-end approach for influence maximization is proposed, simulation on three real-world networks demonstrate that the proposed approach attains as large influence spread as CELF, and with a higher computational efficiency.