Exposure, Skepticism, and Switching: A Behavioral Model of Narrative Diffusion

This study presents a computational framework for understanding how competing narratives spread on social media, using Brazil’s 2022–2023 post-election protests as a case study. We extend the SEIZ epidemiological model to incorporate key behavioral dynamics—exposure, skepticism, resistance, and narrative switching—that influence how narratives evolve in contested digital spaces. Drawing on over half a million posts across Instagram, X (formerly Twitter), and YouTube, we apply physics-informed neural networks (PINNs) to estimate diffusion parameters directly from observed online behavior. Our results show that exposure drives reach, switching reshapes alignment, skepticism slows diffusion, and entrenchment reinforces ideological divides. By capturing these dynamics, our framework offers a behaviorally grounded, data-driven approach to studying narrative competition. The model not only advances diffusion theory but also provides insights for policymakers, platform designers, and researchers seeking to better understand and respond to the spread of divisive or manipulative content online.