<![CDATA[This study provides a comprehensive review of Explainable AI (XAI) applications in fake news detection, addressing the critical "black-box" nature of deep learning models used for misinformation classification. We systematically analyze various interpretability techniques, categorized into ante-hoc and post-hoc methods, applied to neural architectures such as CNNs, RNNs, and Transformers. The study evaluates how these techniques extract linguistic, social context, and visual features to justify classification outcomes. The findings reveal that while attention mechanisms and gradient-based explanations improve transparency, there remains a significant trade-off between model complexity and explanatory clarity. The discussion highlights the challenges of "explanation consistency" and the susceptibility of interpretability tools to adversarial attacks. We conclude that integrating XAI is essential for fostering user trust and regulatory compliance. Future research should prioritize human-centric evaluations to ensure that AI-generated explanations are cognitively accessible to non-expert end-users.]]>
Copyrights © 2024
