Social sentiment early warning system integrating transformers and explainable SHAP values
Yutan Wang · International Journal of Information and Communication Technology · 2026
In recent years, social media has become a key medium for public emotion and thought dynamics, making its sentiment analysis crucial for event prediction.However, while mainstream deep learning models achieve accurate predictions, their 'black box' decision process hampers reliable warning.This paper thus innovatively integrates the powerful transformer model with interpretable Shapley additive explanations values to construct a social sentiment warning framework with both high accuracy and transparency.Experiments on public datasets show the method's comprehensive warning performance significantly outperforms traditional models: area under the curve reaches 0.872, which is approximately 7.5% higher than the classic long short-term memory model, overall warning accuracy rises to 85.6%, and the false alarm rate drops by nearly 12%.This provides an effective solution for reliable and interpretable automated social sentiment perception and early risk warning.