Comprehensive Fake News Detection using Hybrid BERT and Rule-Based Analysis

Abstract

The misinformation is diffused extensively via Internet forms big barriers to primary-retaining information credibility and productive public communication. We propose a new dual component approach which is innovative combines neural architecture based on BERT with customized rule-based analysis. None of the above is complete, as this paper aims to solve the fake news identification problem using a driven analysis. There are 47 deception markers with our framework, as well 15 authenticity indications using a weighted system: Pfinal=0.3 ×Prules+ 0.7 ×PBERTwhich is 91% classification precision and yet processing speeds of less than 400 milliseconds. The design has specialized domain enhancement, real-time web-based assessment, reactivity and real-time user interaction-based learning. Testing across a critically collected collection of 100 various samples indicates improved performance relative to the existing methodologies, and including Athletic content 96 percent precision and 7.8 percent calibration deviation in confidence assessment. Our strategy addresses basic weaknesses in current detection models through providing explainable results, real-time processing speeds and self-bettering algorithms, determining its practicability in the real world information validation applications.