The proliferation of fake news on digital platforms poses a growing threat to democratic integrity, public trust, and social stability. Addressing the limitations of existing detection models, particularly their opacity and computational overhead, this study proposes BiLSTM-LIME, a hybrid framework integrating Bidirectional Long Short-Term Memory (BiLSTM) networks with Local Interpretable Model Agnostic Explanations (LIME) for interpretable and efficient fake news detection. The model employs pre-trained GloVe embeddings to capture semantic and syntactic dependencies, while LIME provides token level explainability, enhancing transparency in model decisions. A standardized English corpus with a refined preprocessing pipeline was developed to ensure robust and reproducible evaluation. Experimental results demonstrate that the proposed BiLSTM-LIME model achieves 97.21% accuracy and an F1 score of 0.97, outperforming several state of the art transformer based and multimodal approaches while maintaining a significantly lower computational cost. The framework establishes a balance between performance and interpretability, offering a scalable, transparent, and resource efficient solution for real world fake news detection.
Sneha et al. (Sat,) studied this question.