The rapid dissemination of fake news in the digital age has raised concernsregarding the authenticity and credibility of online information. In this study, we propose an innovative approach for detecting fake news by combining the power of RoBERTa, a state-of-the-art language model, with the Firefly Optimization Algorithm (FOA) and Long Short-Term Memory (LSTM) neural networks. Our model leverages RoBERTa’s contextual understanding of text to extract informative features from news articles, including content, user engagement, and source characteristics. The FOA is employed to optimize the selection of these features, enhancing the model’s ability to discern between genuine and deceptive news. The selected features are then used as inputs to an LSTM network, which learns the temporal dependencies within the data for accurate classification. To validate the effectiveness of our approach, we conduct experiments using benchmark datasets such as ISOT and fakenewsnet specifically designed for fake news detection tasks. The experimental results demonstrate that our proposed model outperforms existing methods in terms of accuracy and other performance metrics. By integrating RoBERTa, FOA, and LSTM, our model effectively combines contextual understanding, feature selection optimization, and sequence modeling, enabling accurate identification of fake news articles. This research contributes to the advancement of fake news detection techniques and offers a promising solution to address the challenges associated with online misinformation, ultimately fostering trust and reliability in the digital information landscape.