An intelligent deep learning and natural language processing framework for fake news detection and text classification in social media forwarded messages
Abstract
Social media platforms such as WhatsApp, Facebook, Twitter and Instagram have become primary channels for news dissemination, but they also allow false information to spread quickly and largely unverified, fueling confusion, fear and mistrust. The volume of forwarded content makes manual verification impractical, so automated detection is needed. This paper presents a reproducible NLP pipeline for fake news detection and text classification, built in five stages: dataset description, data pre-processing, feature extraction, model selection and training, and model evaluation. The Kaggle “Fake and Real News” dataset was used, with Reuters articles as real news and articles from sources known for hoaxes or hyperpartisan content as fake news. A subset of 5,000 articles was used for experiments. Text was cleaned and normalized through tokenization, stop-word removal and stemming, then converted into numerical vectors using Term Frequency–Inverse Document Frequency (TF-IDF), which quantifies the importance of each word within the corpus. A Logistic Regression classifier was selected for its interpretability and strong baseline performance and was trained and evaluated on separate training and test partitions. The model achieved a training accuracy of 98.78% and a test accuracy of 97.00%, showing that a simple supervised approach combined with effective pre-processing can distinguish real from fake news reliably. The dataset consists of long-form news articles rather than short, forwarded messages, so results on social media forwards may differ, and validation on social-media datasets is left for future work. The trained model can classify pasted text as real or fake with an associated probability, and it is designed to be deployed as a web service so users can check messages quickly. Future work includes evaluation on forwarded-message data, multilingual detection, integration with live social media feeds, and deep learning extensions such as transformer-based models.