D2V-DDQN: Influence Maximization of Positive Opinions Based on Deep Reinforcement Learning

The problem of maximizing influence aims to select a portion of seed users in social networks, making the diffusion wider of users’ opinions. For a topic, the diffusion width is the number of activated users, and the diffusion depth is the average opinion value of all activated users after a certain period of information diffusion on the social network. This paper proposes an opinion-based independent cascade (OPIC) model to quantify the diffusion width and depth of the topic. Then, we propose a deep reinforcement learning algorithm named D2V-DDQN. This algorithm aims to find a seed node set to maximize the positive influence of the target topic so that there are more positive opinions with a large number of activated nodes under the OPIC diffusion model. The experimental results verify the superiority of the method relative to existing methods.