Authors:
Amrutha Sridhanya Madduri, Soundarya Suresh, G. P. Abhilash, Abdulrahman Magdy Ragab Abdelshakour Azzam, Alaa Mohamed Gaber Hussein, Anass Esam Mohamed Mohamed Elsakka
Addresses:
Department of Computer Science and Engineering, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India. Department of Biomedical Sciences, Zewail City of Science and Technology, 6th of October City, Giza Governorate, Egypt.
Abstract:
The rapid spread of misinformation across digital platforms necessitates the development of accurate and efficient fake news detection systems. This study proposes a context-sensitive hybrid deep learning framework that integrates DistilBERT-based transformer embedding with a Bidirectional Long Short-Term Memory (BiLSTM) network to capture both contextual semantics and sequential dependencies in textual data. The model is trained on the ISOT dataset and evaluated using a comprehensive set of performance and agreement metrics. Experimental results demonstrate that the proposed model achieves high classification performance, with an accuracy of 0.9993, precision of 0.9997, recall of 0.9990, and F1-score of 0.9994. Structural evaluation further confirms the model’s reliability, with an Adjusted Rand Index (ARI) of 0.9974, a Normalised Mutual Information (NMI) of 0.9924, and a Cohen’s Kappa of 0.9987, indicating near-perfect agreement between predicted and actual labels. However, cross-dataset evaluation on the LIAR dataset reveals a significant performance drop, with accuracy of 0.6429, precision of 0.275, recall of 0.003, and F1-score of 0.005, highlighting limitations in domain generalization. An ablation study validates the effectiveness of the hybrid architecture, showing improved performance compared to individual DistilBERT and BiLSTM models, and demonstrates the impact of input sequence length optimization on both accuracy and computational efficiency. Overall, the proposed framework achieves state-of-the-art results in in-domain evaluation while identifying critical challenges in cross-domain robustness, underscoring the need for further research on generalizable fake news detection systems.
Keywords: Detection of Fake News; Deep Learning; Normalized Mutual Information; Natural Language Processing; Text Classification; Transformer Model; AdamW Optimization; Evaluation Metrics.
Received on: 20/02/2025, Revised on: 19/04/2025, Accepted on: 26/06/2025, Published on: 05/06/2026
DOI: 10.64091/ATICL.2026.000330
AVE Trends in Intelligent Computer Letters, 2026 Vol. 2 No. 2 , Pages: 122-137