Berita Debunked: Real-time Fake News Detection and Alert System

BeritaDebunked is an AI-driven near real-time fake news detection and alert system designed to combat misinformation in Malaysia, particularly on platforms such as WhatsApp. The system combines natural language processing and multimodal deep learning by using BERT for textual analysis and BLIP-2 for image–text evaluation. Deployed as a browser extension, it flags suspicious messages and allows continuous model updates through a scalable backend. Evaluation on the Fakeddit benchmark dataset demonstrates that the proposed hybrid architecture achieves an accuracy of (83.3%), with a precision of (82.6%) and an F1-score of (84.9)%. While unimodal text baselines achieved marginally lower raw accuracy (82.9%), the hybrid model demonstrates superior robustness in detecting multimodal context mismatches. The system demonstrates real-time capability with an average inference latency of 56.42 ms. By enabling timely detection and user-friendly alerts, BeritaDebunked aims to support digital literacy efforts, reduce the spread of misinformation, and contribute to Sustainable Development Goal 16 by strengthening information integrity.