A Rule and Evidence Enhanced Framework to Dynamic Suppression of Health Misinformation Propagation

Abstract

Social networks are vital for public health information access, but they also accelerate the spread of high-risk or false content. Existing moderation approaches lack the precision and adaptability. We propose a rule and evidence enhanced risk assessment and risk-based dynamic communication decision-making framework to suppress the spread of health misinformation. By constructing a hierarchical evidence knowledge base, it leverages a knowledge-augmented large language model (LLM) to assess information risk along three fine-grained dimensions: logical consistency, source reliability, and semantic ambiguity. Propagation permissions of each node are dynamically adjusted by integrating risk score with node credibility and influence. Experimental results demonstrate that, compared to a general-purpose LLM, our method improves accuracy in identifying high-risk content and achieves a lower false-negative rate. In heterogeneous social network simulations, our framework achieves near-complete suppression of harmful misinformation without inadvertently blocking benign information. This work presents a computable, and explainable solution for balancing the free diffusion of health information and the prevention and control of risky information.