Identification of Rumor Refuters Based on an Explainable Machine Learning Framework

Facing of the existing reception dilemma of rumor refuting information, crowd identification and feature analysis should be realized through big data analysis. Our approach aims to accurately identify rumor refuters and interpret predictions. Initially, we compare six machine learning models, with results demonstrating the superiority of eXtreme Gradient Boosting (XGBoost) over other advanced models. Subsequently, we introduce Shapley additive explanations (SHAP) to interpret the predictions of complex machine learning models and assess the importance of various features. Our findings underscore that utilizing XGBoost alongside the SHAP approach can offer decision support, enhancing the effectiveness of rumor governance.