Advancements in Deep Learning for Fake News Detection: A Comprehensive Review of Techniques, Datasets, and Emerging Trends

Fake news on social media has emerged as a major challenge, significantly affecting politics, public health, social trust, and economic stability. The rapid expansion of digital platforms has enabled the widespread circulation of misinformation, exposing the limitations of traditional fake news detection methods such as rule-based systems and conventional machine learning techniques. These approaches struggle to manage the scale, semantic diversity, structural variation, and multimodal nature of contemporary fake news.
Deep Learning (DL) has proven to be a powerful alternative by enabling automated feature learning and effective processing of large volumes of unstructured data. Advanced DL models, particularly transformer-based architectures and self-attention mechanisms, have demonstrated superior performance in capturing contextual and social information. Techniques including Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) networks, Graph Neural Networks (GNNs), pruned transformer models such as BERT, and hybrid frameworks have shown promising results.
This paper reviews recent advancements in deep learning-based fake news detection, highlighting key models, datasets, evaluation metrics, ethical concerns, multimodal misinformation, and the growing role of Explainable AI (XAI) to guide future research toward accurate and transparent systems.