What Is Said, Who Says It, and How It Spreads: A Socio-Semantic Graph Framework for Fake News Detection

The proliferation of misinformation on social media poses a critical challenge. Existing detection approaches often rely on isolated signals like content, propagation structure, or user credibility. However, these signals are ambiguous in isolation: propagation patterns require semantic interpretation, while source credibility must be contextualized by the specific claims being made. Efficient veracity detection therefore lies in modeling their complex interplay. We address this with a unified socio-semantic graph framework that jointly models what is said, who says it, and how it spreads. Our model represents conversational cascades as attributed graphs, enriched with semantic embeddings and a novel behavioral reputation score that penalizes visibility amplification. Veracity is assessed through the credibility-weighted consensus that emerges from these dynamics. Experiments on the RumourEval 2019 benchmark demonstrate our approach's effectiveness, achieving a macro F1-score of 63.55\%, which compares favorably to state-of-the-art methods. Our work lays the groundwork for scalable credibility analysis in large-scale environments.