Emotion-aware dual cross-attentive neural network with label fusion for stance detection in misinformative social media content

Abstract

Stance detection determines a user’s opinion toward a particular target or statement. The task helps analyze underlying biases in shared information and combat misinformation. Social media generates massive amounts of user-generated content (UGC). This content often conveys implicit opinions which contribute to the spread of misinformation. We propose a Stance Prediction through a Label-fused dual cross-Attentive Emotion-aware neural Network (SPLAENet) in misinformative social media user-generated content. It uses a dual cross-attention mechanism. This mechanism focuses on relevant parts of source text in the context of reply text, and vice versa. We incorporate emotions to distinguish between stance categories. Emotional alignment or divergence between texts helps separate different stances. We also employ label fusion that uses distance-metric learning to align extracted features with stance labels. This technique improves the method’s ability to accurately distinguish between stances. Extensive experiments demonstrate the significant improvements achieved by SPLAENet over existing state-of-the-art methods. SPLAENet improves over existing methods across three datasets. On RumourEval dataset, our method shows an average gain of 8.92% in accuracy and 17.36% in F1-score. On the SemEval dataset, it gains 7.02% in accuracy and 10.92% in F1-score. On the P-stance dataset, it shows average gains of 10.03% in accuracy and 11.18% in F1-score. These results validate the effectiveness of the proposed method for stance detection in the context of misinformative social media content.