Beyond the hotspots: Identifying public opinion blind spots and agenda-setting dynamics in crisis information generation

During sudden crises, a structural mismatch between social media attention and governance value often marginalizes key crisis issues, creating public opinion blind spots that undermine effective emergency management. Using the 2021 Henan rainstorm as a case study ( N = 40,160), this study proposes an integrated framework to identify public opinion blind spots and examine their agenda-setting dynamics. Methodologically, we combine grounded theory with Large Language Model (LLM) analysis for theme classification (92.7% accuracy), apply refined Importance-Performance Analysis (IPA) to identify blind spots, and use Vector Autoregression (VAR) models, Granger causality tests, and impulse response functions (IRFs) to examine agenda-setting dynamics. Bootstrap estimation further assesses the association between blind spots and agenda-setting effects. Results show that blind spots vary dynamically across the 4R phases, with Prevention Education, Disaster Causation, and Review and Improvement recurring most frequently. Bidirectional agenda-setting effects can emerge at a 1-hour resolution, and stronger agenda guidance from authoritative users is associated with blind-spot mitigation. Ultimately, these findings offer practical implications for optimizing crisis information supply and strengthening social resilience.