Paper Details

Name: Avinash Bhat & Sangeetha J
Country: India

Year: 2026
Volume: Volume No: 13, January, Year: 2026 (Special Issue)
Page Number: 463-470
DOI: https://doi.org/10.5281/zenodo.19050823

ABSTRACT
The integrity of information and the public's confidence are under serious attack due to the dissemination of false news on social media. Leveraging transformer-based language models more specifically, the Distil BERTbase- uncased and RoBERTa-base models. this study provides a dependable and efficient approach for identifying fake news. The light structures of these models as well. the ability to retain important contextual information render them suitable for high-performance text classification tasks. To manage the complexity of detecting false news, training and testing were done on large datasets incorporating various types of news content, user interaction, and location information. RoBERTa achieved a competitive performance with an accuracy of 89% and an F1 score of 92%, while Distil BERT attained an accuracy of 86% and an F1 score of 91%. In terms of efficiency and computational cost, both models surpassed traditional machine learning methods. Also, the incorporation of extra social environment features which were inspired by advances in the discipline—was needed to maximize model predictions These findings contribute to the growing body of research indicating that massive pre-trained language models can be used to combat disinformation. For further enhance detection abilities on social media platforms, future studies might explore real time optimization techniques and multi-class classification situations.

Keywords— Fake News Detection, RoBERTa, Distil BERT- base-uncased, Natural Language Processing, Transformer Models, Text Classification, Social Media Misinformation, Machine Learning, F1 Score, Accuracy, Binary Classification