Real-Time Fake News Detection using DeBERTa-V3 and Transformer

Abstract

Fake news dissemination through digital and social media has grown into a real threat, as it distorts public opinion, decision-making, and confidence in institutions. Manual fact-checking is time-consuming and slow, hence inappropriate for a real-time environment where the misinformation spreads within minutes. This study proposes a real-time fake news detection framework that incorporates the pre-trained DeBERTa V3 transformer model for text classification, fact-checking based on Wikipedia, and sentiment analysis to provide explainable and transparent outputs. The system is trained and tested by the LIAR factchecking dataset of short political statements labeled, and its deployment is performed via an intuitive interface using Gradio so that users can easily input the news text and get classifications, sentiments, and evidence that support those sentiments in real-time. The experimental results indicated that the proposed system can effectively classify fake from real news, while providing interpretable justifications, hence improving users' trust and promoting responsible consumption of online information. The proposed approach is appropriate for embedding into media monitoring tools, browser extensions, and learning environments to mitigate the effects of misinformation.