[Objective] Extract external stimuli and cognitive evaluation indicators from users, construct a model of extreme emotional influencing factors for Weibo users under negative corporate events, and use SHAP to explain the impact of each characteristic variable. [Methods] Based on the cognitive affective theory, social impact theory, emotional evaluation model, LDA model, and Grounded theory, this study determines the external stimulus and cognitive evaluation indicators, and uses the characteristic variables contained in the two indicators as inputs and extreme affective variables as outputs to build an extreme affective impact factor model. Compare the performance of four classification models and visualize the optimal model using the SHAP model. [Results] The results showed that the cognitive evaluation dimension extracted 7 feature variables; The accuracy of the LGBM model reaches 0.88, Precision reaches 0.90, and F1 score reaches 0.93, which is superior to other comparative models; From the perspective of the impact of characteristic variables on the extreme emotions of Weibo users, the cognitive evaluation dimension is generally higher than the external stimulus dimension, and the impact of each characteristic variable varies. [Limitations] It is necessary to explore more influencing factors and a wider range of negative event types in enterprises, and the algorithm complexity needs to be improved. [Conclusions] The model proposed in this study optimized the rooting encoding process, visualizing the degree, direction, magnitude, and mode of influence of each feature variable on extreme emotions, providing a theoretical basis for enterprises to solve the problem of negative online word-of-mouth.