Explainable Deep Learning for False Information Identification: An Argumentation Theory Approach.

The article focuses on developing an automated false information identification (FII) framework grounded in Toulmin's model of argumentation and structural balance theory (SBT) to improve both accuracy and explainability in detecting false claims. The proposed method, called G-FINDER, constructs signed word networks from claim-evidence pairs to assess their semantic and syntactic consistency, enabling machine learning models to better verify claim veracity while providing human-understandable explanations (SBTX). Experiments on real-world Twitter rumor data and crowdsourced fact-checking datasets demonstrate that G-FINDER enhances the performance of various baseline models and that its explanation approach improves human task accuracy, trust in AI, and confidence in decision making compared to attention-based explanations. The study highlights the importance of integrating linguistic and psychological theories into AI for false information detection and suggests future research directions including evidence retrieval, handling machine-generated false information, and dynamic evidence scenarios.