International Journal of Advanced Research in Computer and Communication Engineering

Abstract: The rising proliferation of fake news on digital platforms has become a serious social and technological challenge. Fake news, which is defined as false or misleading information that is passed off as reality, has the potential to sway elections, exacerbate divisiveness, and jeopardize public health. Existing detection techniques, which include deep learning, transformer-based architectures, and handmade machine learning classifiers, perform well in benchmark scenarios but have three enduring issues: lack of explainability, domain shift, and adversarial paraphrase.

This study combines explainability modules, cross-domain evaluation, and adversarial data augmentation to present a strong NLP-driven framework for false news identification. The pipeline compares traditional models such as Support Vector Machines with newer designs like BiLSTM, BERT, and RoBERTa, coupled with a hybrid ensemble (BERT + SVM). Several benchmark datasets from the political, health, and entertainment domains—LIAR, FakeNewsNet, BuzzFeed, FEVER, Weibo, and Hinglish—were evaluated.The results show that hybrid ensembles achieve improved robustness against adversarial attacks and temporal drift, whereas transformer-based models continuously outperform previous methods. The results emphasize how crucial it is to integrate factual, stylistic, and semantic elements in order to create robust and interpretable fake news detection algorithms for practical uses.

Downloads: | DOI: 10.17148/IJARCCE.2025.14843