Transforming Fake News Detection: Leveraging DistilBERT Models for Enhanced Accuracy

In today’s digital world, social media is a vital part of our lives, the widespread occurrence of deceptive information on social media sites poses a considerable obstacle to effectively recognizing fake news. BERT, which stands for Bidirectional Encoder Representations from Transformers, is an advanced transformer-based model designed to comprehend the context from both directions within the text, which greatly enhances its ability to comprehend various NLP tasks. Moreover, DistilBERT ofers a more streamlined, quicker, and more resource-efficient alternative to BERT, providing approximately 97% of BERT’s effectiveness. This characteristic renders DistilBERT an exemplary option for applications in environments where resources are constrained. We rigorously tested four DistilBERT models and validated our proposed model by conducting extensive experiments on the Jruvika dataset using multiple evaluation metrics. Our best-performing model, DistilBERT base-cased-distilled-squad, achieved an impressive 0.977 accuracy and maintained a consistent F1-Score of 0.97. Our research indicates that the suggested approach significantly enhances the accuracy of identifying false information.