The rise of online social networks has revolutionized communication and has become the primary platform for information sharing. However, this broad accessibility of information has also facilitated the rapid spread of misinformation, emphasizing the critical need for detecting fake news. Traditional machine learning methods, while useful, often require hand-crafted features and focus mainly on text analysis, overlooking the complex, multimodal nature of fake news. Recent advances in graph neural networks (GNNs) have shown promise in addressing these challenges, but existing models still face significant limitations in capturing both local syntactic dependencies and broader contextual information, and in providing explainable results. To address these issues, we present IMGA-Net: inconsistency-aware multiview graph attention network for explainable fake news detection. IMGA-Net leverages both the sequential context and syntactic structure of tweet content through a bi-graph attention network (BiGAT). By constructing a propagation graph network and a propagation graph convolutional network, IMGA-Net captures the propagation of retweets and identifies key contributors to the spread of fake news. Furthermore, the model incorporates an inconsistency learning layer to detect semantic incon- sistencies across multiple views, enhancing its ability to identify fake news. IMGA-Net also emphasizes explainability, providing insights into the evidence supporting its decisions. Unlike previous studies that leverage GATs primarily for syntactic or sentiment analysis, our proposed IMGA-Net offers a multiview, inconsistency-aware framework that unifies propagation dynamics, semantic representations, and syntactic dependencies through a cross-view consistency alignment mechanism. This design enables IMGA-Net to effectively resolve conflicting cues across different information views, setting it apart from existing approaches(https://github.com/maryamdolati/IMGA-Net.git).
