Abstract
Global public emergencies occur frequently, and the risk of Internet public opinion crises in such contexts is gradually increasing. In the dual context of risk society and network society, effectively identifying and assessing Internet public opinion risks on public emergencies poses challenges to the efficiency and response speed of public crisis management. This paper innovatively proposes an Internet public opinion risk identification and assessment method for public emergencies, integrating association rule mining with Bayesian network (BN). The core innovation lies in designing an improved scheme based on the CBA (Classification Based on Associations) algorithm to overcome the limitation of traditional association rule mining in handling non-Boolean data, thereby effectively extracting the correlations among public opinion risk factors to optimize the topological structure of the BN. Building upon this foundation, we construct a BN model with strong interpretability to identify the public opinion key risk factors and key risk chains of different risk levels, as well as to evaluate the public opinion situation. Empirical results demonstrate that, compared with the traditional BN assessment model based on expert experience, the BN model incorporating association rules achieves a 14.4% increase in assessment accuracy and exhibits more pronounced advantages in performance metrics such as precision, recall, specificity, and F-measure. The proposal of this innovative method not only enhances the accuracy of public opinion risk assessment but also provides a new perspective for data-driven identification of key risk factors and research on their complex interactions. Furthermore, it provides an interpretable and computationally efficient decision support tool for public opinion crisis management.