Abstract for: Modeling the Evolution of Public Opinion Risks Based on the Integration of Causal Inference and System Dynamics

The rapid growth of social media has introduced complex, non-linear mechanisms to online public opinion dissemination, posing significant challenges to traditional static analysis and public governance. This study addresses this gap by proposing a framework that integrates causal discovery with system dynamics to reveal the dynamic propagation mechanisms of online public opinion risks. By incorporating information entropy and multimodal coverage, the traditional indicator system is expanded, and the PC-LiNGAM algorithm is employed to identify dynamic causal relationships. Based on these insights, a system dynamics model is constructed, featuring key variables such as government reports, public emotions, and the Public Opinion Index with Dynamic Weights (PODW). Building upon the classic SIR model, an improved multi-agent system dynamics model is developed to integrate multi-state emotional transitions and dynamic feature regulation. Simulation results indicate that information entropy and multimodal coverage are critical drivers of risk escalation. Furthermore, the study demonstrates that timely authoritative government interventions can effectively reduce uncertainty, steer public sentiment, and curb risk propagation. These findings provide a scientific basis for risk early warning and suggest a governance shift from passive "blocking" to active "guidance" strategies.