Deep reinforcement learning-based approach to tackle positive influence maximization in signed social networks

The Influence Maximization problem has garnered significant research interest since its introduction. In 2014, the problem was further extended to include signed social networks, resulting in the positive influence maximization problem and negative influence maximization problem. However, current solutions mainly rely on greedy algorithms that suffer from the Inefficient shortcoming, which do not leverage deep reinforcement learning. This paper introduces a novel approach to maximize the number of positively activated nodes by applying deep reinforcement learning to signed networks. Specifically, we extend SDGNN model for network representation learning and design a DQN-based seed node selection algorithm. The extensive experimental results on four real-world networks demonstrate our proposed model outperforms greedy algorithm and CELF algorithm, in terms of both time efficiency and influence spread quality. To our knowledge, this work is the first to leverage deep reinforcement learning to solve influence maximization problem in signed social networks.