Graph-Augmented Language Model Framework for Health Misinformation Detection

In today’s digital world, the spread of fake news is a growing concern—especially in the medical field, where misinformation about diseases, treatments, and vaccines can have serious consequences for public health. Misleading medical content spreads quickly, causing confusion, undermining trust in healthcare, and influencing critical decisions. To address this challenge, our study introduces an automated system for detecting medical misinformation using a combination of advanced deep learning and graph-based techniques. We evaluate several models, including BERT, GPT-Neo, Graph Neural Networks (GNN), and a hybrid GPT-GNN approach, to classify health-related news articles as real or fake. Our analysis is based on a well-rounded dataset of 28,945 records, drawn from multiple trusted sources such as FakeHealth, MedHub, diabetes-related misinformation datasets and COVID-19 collections. The dataset includes 14,838 real and 14,107 fake news samples. The proposed hybrid GPT-GNN model achieves 96.1% accuracy with statistical significance (p < 0.001) across multiple validation runs, demonstrating superior performance compared to recent baselines including GraphBERT and RoBERTa-GNN. To improve model performance, we apply comprehensive preprocessing steps like tokenization, stopword removal, and vectorization. The results are promising: the hybrid GPT-GNN model outperforms individual models, achieving higher accuracy in detecting false information. By blending the contextual understanding of transformer models with the relational insights offered by graph-based learning, our approach provides a scalable and reliable solution for identifying medical misinformation and ultimately, for helping people make more informed healthcare decisions.