This study takes the dual dimensions of user stance and sentiment as an entry point, with the aim of mining the sentiment and stance elements in comments of misinformation and investigating whether combining comment environment features with content features can improve the effectiveness of detection models.
This study proposes a feature set for misinformation detection, including content features and comment environmental features. Through statistical tests and regression analysis, features that facilitate the detection of misinformation are determined. A series of machine learning (ML)-based and deep learning (DL)-based models are employed to assess whether comment environment features can help improve model performance. Finally, Shapley Additive Explanations (SHAP) are applied to interpret feature contributions.
The experimental results indicate that there are significant differences between misinformation and authentic information, not only in content but also in the comment environment. And the introduction of comment environment features effectively enhances the performance of detection models.
This study proposes a new misinformation detection approach from the perspective of combining user stance and sentiment. This approach has been validated through the application of ML-based models, DL-based models and further supported by SHAP analysis, offering novel insights into the governance of misinformation on social media platforms.
