Evidence-Aware Fake News Detection Using External Knowledge and Neural Networks

Abstract

The proliferation of fake news on digital platforms poses a significant threat to public opinion, decision-making, and social stability. Traditional fake news detection approaches rely primarily on textual features of claims, often ignoring external evidence, which limits their reliability and interpretability. This study proposes an evidence-aware fake news detection approach that incorporates both claim text and fact-checking justification as external knowledge. Two transformer-based models were developed using the LIAR dataset: a baseline claim-only model and an evidence-aware model built on the DeBERTa-v3-small architecture. Both models were trained and evaluated using standard classification metrics, including accuracy, precision, recall, and F1-score. Experimental results indicate that, while both models achieved similar accuracy (approximately 60%), the evidence-aware model significantly improved the F1-score and recall for the fake news class, increasing recall from 21.2% to 38.3%. These findings demonstrate that incorporating external evidence enhances the model’s ability to detect fake news and yields more balanced classification performance. The results highlight the limitations of relying solely on textual patterns and emphasize the importance of integrating external knowledge for more reliable and effective fake news detection systems. Both models use a lightweight, resource-efficient transformer (44M parameters) trained on a single consumer-grade GPU, and results are reported on the LIAR benchmark only; broader interpretability analysis and cross-dataset evaluation are identified as directions for future work.