From Accuracy to Trustworthiness: A Survey of Explainable AI Approaches for Fake News Detection Using Transformers and Graph Neural Networks

The rapid proliferation of misinformation across digital platforms has emerged as a significant challenge, influencing social communication, public opinion, and decision-making processes. Recent advances in artificial intelligence (AI), particularly Transformer-based architectures and graph neural networks (GNNs), have substantially improved the performance of fake news detection systems. However, achieving high classification accuracy alone is insufficient for developing reliable, transparent, and trustworthy AI-based solutions. This paper presents a comprehensive review of recent AI-driven approaches for fake news detection, emphasizing explainability, interpretability, transparency, and trustworthiness. The evolution of detection methods from traditional machine learning algorithms to advanced deep learning architectures, including Transformers, graph-based models, and multimodal learning frameworks, is systematically analyzed. Furthermore, this review investigates the role of explainable artificial intelligence (XAI) techniques, including attention visualization, LIME, and SHAP, in enhancing the interpretability of automated decisions. Existing challenges related to model bias, data quality, adversarial attacks, and real-world deployment are discussed, and future research directions toward robust and human-centered fake news detection systems are highlighted. This study provides researchers and practitioners with a structured overview of current methodologies and emerging opportunities for developing trustworthy AI solutions against misinformation.

کلیدواژه ها:

Fake News Detection ، Explainable Artificial Intelligence ، Transformers ، Graph Neural Networks ، Natural Language Processing ، Trustworthy AI