VeriFactAI: A Hybrid Deep Learning Framework for Accurate Text-based Fake News Detection Using RoBERTa and BiLSTM

In brief, the sudden viral spread of false news through digital channels turns out to be a new challenge for society by stealing people's trust in reliable sources of information. The existing detection techniques are mainly based on shallow feature extraction and conventional machine learning models, failing to capture misinformation's complex and dynamic nature. Furthermore, these methods are not scalable and adaptable when handling content in its diverse and multilingual nature. Hence, a hybrid deep learning framework known as VeriFactAI was introduced to resolve this. It uses advanced natural language processing and optimized training strategies to break the row of accuracy in detecting fake news. This model combines the richness of a contextual embedding transformer, RoBERTa, and a BiLSTM, making the system deeply understand the dependency between text. The system begins with high-end text preprocessing, yielding embeddings passed through the hybrid model for robust classification. It also aims to accelerate convergence speed and diminish the possibilities of overfitting during training by introducing a novel optimizer, AdamTuner, an upgrade of the traditional AdamW and fine-tuned hyperparameters that enhance convergence and generalization. The proposed system demonstrates excellent accuracy, precision, and recall performance on benchmark datasets. It is scalable and very efficient towards offering a solution against fake news by being an effective tool in the fight against the spread of misinformation. It benefits media platforms, researchers, and policymakers in preventing the adverse effects of fake news.