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

2025-05-27

Engineering applications of artificial intelligence

AI Summary

To address the challenges of ambiguous user stances and insufficient coupling between sentiment and logical reasoning in misinformation propagation, this paper proposes a sentiment-aware stance detection method. Methodologically, we design a BERT-based dual-channel encoder, incorporating an emotion-gated cross-attention mechanism and a sentiment–stance label fusion strategy to enable fine-grained sentiment-guided stance classification; we further introduce a multi-label consistency fusion module and adversarial emotion-augmented training to enhance model robustness. On FakeNewsNet and PStance benchmarks, our approach achieves absolute F1-score improvements of 4.2% and 3.8%, respectively, and reduces sentiment-related error rates by 21%, significantly improving detection accuracy for ironic, implicit, and polarized stances. The core contribution lies in the first joint optimization of dynamic emotion modeling and argument structure, overcoming the longstanding limitation in conventional stance detection—namely, the semantic and inferential decoupling between sentiment and logical reasoning.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Knowledge Representation and Reasoning: Argumentation
• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Sentiment analysis and opinion mining
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Problem

• Detect stance in misinformative social media content
• Leverage emotion and label fusion for stance prediction
• Improve accuracy in distinguishing polarized user opinions

Innovation

• Dual cross-attention mechanism captures inter-intra relationships
• Emotion-aware network distinguishes stance categories effectively
• Label fusion aligns features with stance labels accurately