Tracking evolving communities in fake news cascades using temporal graphs.

Misinformation proliferates on social media platforms owing to both static and dynamic user populations, where the set of active users and their interactions evolve over time. The development, amalgamation, or disintegration of communities throughout an information cascade complicates the longitudinal tracking of these communities. Numerous contemporary methodologies either neglect temporal factors or employ static clustering techniques, which do not accommodate dynamic coordination. We propose TIDE-MARK, a methodology designed to identify communities inside fake news cascades that exhibit consistency in both structure and temporal dynamics. The methodology encompasses node embeddings via temporal graph neural networks, prototype-driven clustering, Markov modeling of community transitions, and reinforcement-based refinement. The unified design facilitates consistent and comprehensible community trajectories. Three empirical datasets pertaining to political, entertainment, and health-related fake news are utilized to evaluate TIDE-MARK. The databases include PolitiFact, GossipCop, and ReCOVery. Our model surpasses robust baselines regarding structural (modularity, conductance) and temporal (adjusted Rand index) measures, supported by consistent effect sizes. Structural research indicates that real news spreads through more scattered and less organized communities, while false news propagates through more stable and well interconnected communities. Our objective is to assess the viability of interventions by simulating a structure-aware approach that targets important users in nascent communities. The substantial reduction in cascade modularity and spread demonstrated in the results demonstrates the potential viability of content-neutral mitigation techniques. TIDE-MARK offers a structure-aware framework for real-time fake news monitoring, emphasizing network-based strategy signals over textual analysis. It establishes a foundation for innovative methods of dynamic community monitoring inside complex social systems and features an interpretable architecture that enables ethical application.