Abstract: Based on the overconfidence and evolutionary game, the multi-agent simulation model is constructed to study the public sentiment diffusion rules of online social networks The online social network public sentiment diffusion model is constructed from the perspective of evolutionary game; combined with the theory of overconfidence, three overconfident scenarios are designed: benefit overconfidence, cost overconfidence, and dual overconfidence of cost and benefit Then historical information learning rules of their own and neighbor is designed to achieve multi-agent simulation of public sentiment diffusion under different overconfident scenarios The results show: 1) The stronger the benefit overconfidence is, the faster the convergence speed of public sentiment diffusion is, and the faster the user group reaches a stable state 2) Increasing the cost overconfidence will result in the reduction of adoption proportion, and transformation of group strategy from adoption dominance to tit for tat 3) Compared with benefit overconfidence, cost overconfidence is more conducive to improving the stability of public sentiment diffusion This research attempts to propose an integration framework for behavioral theory and simulation methods in order to provide new ideas for public sentiment diffusion research of online social networks

Key words: public sentiment diffusion, multi-agent simulation;