POSIM: A Multi-Agent Simulation Framework for Social Media Public Opinion Evolution and Governance

Abstract

Modeling social media public opinion evolution is essential for governance decision-making. Traditional epidemic models and rule-based agent-based models (ABMs) fail to capture the cognitive processes and adaptive behaviors of real users. Recent large language model (LLM)-based social simulations can reproduce group-level phenomena like polarization and conformity, yet remain unable to recreate the irrational interactions and multi-phase dynamics of real public opinion events. We present POSIM (Public Opinion Simulator), a multi-agent simulation framework for social media public opinion evolution and governance. POSIM integrates LLM-driven agents with a Belief--Desire--Intention (BDI) cognitive architecture that accounts for irrational factors, places them in a virtual social media environment with social networks and recommendation mechanisms, and drives temporal dynamics through a Hawkes point process engine that captures the co-evolution of agents and the environment across event phases. To validate the framework, we collect real-world public opinion datasets from the Weibo platform covering the full interaction chain of users. Experiments show that POSIM successfully reproduces key characteristics of public opinion evolution from individual mechanisms to collective phenomena, and its effectiveness is further supported by multiple statistical metrics. Building on POSIM, governance-oriented guidance and intervention experiments uncover a counterintuitive empathy paradox: empathetic guidance deepens negative sentiment instead of easing it under certain conditions, offering new insights for governance strategy design. These results demonstrate that the proposed framework can fully serve as a computational experimentation platform for proactive strategy evaluation and evidence-based governance. All source code is available at https://github.com/DeepCogLab/posim/.

Figures (7)

• Figure 1: Overall architecture of POSIM. (1) The initialization module maps real user data and post data to agent profiles and environment initial states. (2) The Social-BDI agent architecture implements the cognitive pipeline: Perception $\to$ Belief $\to$ Desire $\to$ Intention $\to$ Action. (3) The simulation environment consists of a virtual social media platform and a temporal engine. The platform provides personalized content recommendations, and the temporal engine governs agent activation rhythms based on the Hawkes self-exciting point process. Agent behavioral outputs feed back into the environment to drive subsequent activations, forming a continuous interaction loop. (4) Strategy evaluation uses this loop as its simulator core, with an intervenor for injecting governance strategies and an evaluator for quantitative analysis, supporting counterfactual assessment.
• Figure 2: Simulated public opinion lifecycle, showing posting volume (bars, left axis) and cumulative posting percentage S-curve (solid line, right axis). $E_1$--$E_7$ mark exogenous event injection points.
• Figure 3: Behavioral heterogeneity across four agent types. (a) Emotional intensity temporal evolution; (b) Content length distribution; (c) Multi-dimensional behavioral profile radar chart.
• Figure 4: Emotion polarization index (PI) evolution over simulation rounds. Solid line: PI mean; shading: 90% bootstrap confidence interval; dashed lines: early and late stage means.
• Figure 5: Topological properties of the interaction network. (a) Degree distribution power-law fit $P(k) \sim k^{-\gamma}$; (b) Cascade size CCDF $P(X \geq s) \sim s^{-\alpha}$.