Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{209638,
        author = {Sandeep Yadav and Saud Haseb and Shiv Shankar Yadav and Vaishali Singh},
        title = {Fake News Detection Using BERT-Based Language Model-With Explainable AI},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {5},
        pages = {2827-2837},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=209638},
        abstract = {social media has become one of the fastest means of spreading false information, thus there is an urgent need to develop automatic fact-checking mechanisms. The prior fake news detection systems were largely based upon manually engineered lexical features and classical machine learning classification techniques. Recent studies in deep learning have shown that deep learning models are capable of developing representations. 
However, deep learning model are limited in their ability to understand context and provide meaningful explanations. Many of the early methods for detecting false information used manual extraction of linguistic attributes in conjunction with conventional statistical classifications. We are developing an end-to-end system for detecting fake news. This system combines a fine-tuned version of BERT (Bidirectional Encoder Representations from Transformers), with an Explainable AI Module. 
The proposed system will be applied to the task of binary classification on social media posts and/or online news articles written in English that range from short-to-medium length. We have used WELFake as our main dataset for this study, and we intend to validate our results by comparing them with ISOT. We will also compare our results with those from other fake news datasets such as LIAR. Our methodology involves the following steps: cleaning and normalizing the data; tokenizing the data with BERT WordPiece tokenizer; creating sequences; fine-tuning BERT-base-uncased; performing binary prediction; providing post-hoc explanations of the predictions using LIME and SHAP; and visually presenting token-level importance. As this is a planning document for a future research study, we do not present actual accuracy, precision, recall, F1 score, latency, or statistically significant differences. Instead, we describe the methodology, evaluation protocol, system architecture, and expected outcomes that will guide the empirical evaluation of our proposed framework.},
        keywords = {BERT, Explainable AI, fake news detection, LIME, misinformation, Natural Language Processing, SHAP, text classification, Transformer},
        month = {October},
        }