Consensus achievement strategy of opinion dynamics based on deep reinforcement learning with time constraint
研究了时间约束下如何通过深度强化学习算法提升群体意见的共识达成效率，提出了意见管理框架OMFDRL，适用于重大公共危机等紧急决策场景。
Group opinion often has an important influence on the development and decision-making of major events. However, there are existing two problems with group opinion: (1) As opinions evolve, group opinion may diverge sharply, which is not conducive to obtaining final decision opinion. (2) The evolution of opinions can also cause serious systemic biases in group, which can lead to a final decision that is far from the truth. Hence, this paper deeply investigates two important strategies of consensus boost and opinion guidance in opinion management. Meantime, considering the urgency of some decision-making problems, such as major public crisis events, opinion management process is also subject to time constraint. In this paper, we firstly formalize the minimum adjustment cost consensus boost and opinion guidance with time constraint as Markov decision process because of the intersection and evolution rule of opinions described by opinion dynamics holds Markov property. In this case, the minimum adjustment cost can improve the efficiency of opinion management. We further propose consensus boost algorithm and opinion guidance algorithm based on deep reinforcement learning, which availably mirrors human learning by exploring and receiving feedback from opinion dynamics. Then, by combining the above-mentioned algorithms, we design a new opinion management framework with deep reinforcement learning (OMFDRL). Finally, through comparison experiments, we verify the advantages of our proposed OMFDRL, which can provide managers with more flexible usage conditions.