Detecting and Refining Misinformation Using Domain- Adversarial Graph Neural Network: A Hybrid Method

The swift spread of fake news on social media urges for strong and effective detection system. This study brings out an innovative approach that combines graph convolutional networks with Domain Adversarial Neural Networks (DANN) to capture patterns that distinguish genuine content from false information. Graph Convolutional Network (GCN) are used to examine the complex connections between content and metadata to detect distinguishing patterns between misinformation and legitimate information. Yet, the precision of models created from single domain may be affected by the accuracy of models trained on various domains. To resolve this issue, the proposed approach incorporates DANN to lessen domain disparities and allow the model to generalize across a wide range of social media contexts. To analyse multi-entity interactions and various relationships types, we combine GCN with DANN enabling the model to leverage insights across different domains. By incorporating this enhancement, the hybrid method surpasses traditional methods, achieving a 13% higher F1-score by integrating GCNs and DANN. In comparison to GCN-only and DANN-only models, there is a 12-15% increase in precision and a 10-14% increase in recall, leading to improved scalability and domain adaptability. The integration of attention mechanisms results in enhanced prioritization of important nodes, leading to a notable increase in detection accuracy. Future enhancements focus on adaptive learning, multilingual support and multimodal content analysis enhancing global adaptability and impact.