Deep Reinforcement Learning-based Approach to Tackle Competitive Influence Maximization

Competitive Influence Maximization (CIM) problem studies the competition among multiple parties where each party aims to maximize their profit while competing against other parties. Recently, Reinforcement-Learning based models have been proposed to address the CIM problem. However, such models are unscalable and incapable of handling changes in the network structure. Motivated by the recent success of Deep Reinforcement Learning models and their capability to handle complex problems, we propose a novel Deep Reinforcement learning based framework (DRIM) to address the multi-round competitive influence maximization problem. DRIM framework considers the community structure of the social network for budget allocation and feature extraction with deep Q network in order to reduce the computational time of seed selection. The proposed framework employs the quota-based ε-greedy policy to explore the optimality of influence maximization strategies and budget allocation for each community. Experimental results show that the proposed DRIM framework performs better than the state-of-art algorithms to tackle the multi-round CIM problem.