FakeXE: An Explainable Fake News Detection Using Roberta and LIME

The rise of digital media platforms has led to an increase in the spread of fake news and information, posing an eminent threat to the integrity of online information platforms. Traditional human verification processes are not effective because of the large amount of information being generated at an extraordinary pace, thus making it necessary to have an effective automated system for the detection of fake news. This paper proposes FakeXE, an explainable fake news detection framework that combines the high classification capability of RoBERTa with Local Interpretable Model-agnostic Explanations (LIME) to enhance transparency. The work has been tested on the ISOT Fake News Dataset with the help of traditional machine learning models namely Logistic Regression, Decision Tree, Random Forest, and Gradient Boosting and transformer models like BERT, DistilBERT and RoBERTa. For the aim of analysis, accuracy, precision, recall, and F1-score values have been used. Results show that the transformer models have outperformed the traditional machine learning models with RoBERTa having an accuracy of 99.96% and precision, recall, and $\mathbf{F 1}$-score of more than 99%. However, to cope with the issue of black-box models, the approach of Local Interpretable Model-agnostic Explanations (LIME) is used. It ensures explanations for predictions on the feature level. Thus, it helps to reveal essential words that have a significant impact on classification, thereby increasing the interpretability of the model for users. In summary, it can be said that the presented research proves the efficiency of high-performing deep models integrated with explainable AI.