Social media has become an important channel for monitoring public reactions and supporting risk communication during urban flood events. This study examines emotional dynamics during the 2024 Guilin flood in China using 10,972 Sina Weibo posts collected through keyword-based web crawling. By integrating sentiment classification, topic modeling, and spatial regression, the study analyzes temporal change, thematic variation, and geographic differences in negative sentiment throughout the disaster process. Results reveal a clear four-stage emotional trajectory. During the outbreak phase, the proportion of negative sentiment increased by 10.75 percentage points compared with the latent phase, while extremely negative sentiment increased by 5.88 percentage points. Although negative sentiment gradually declined during the recovery and fluctuation phases, it remained above pre-disaster levels. Emotional intensity also varied substantially across regions. Elevated negativity was concentrated not only in the affected Guangxi region but also in information-centered areas such as Beijing, suggesting that intensive information circulation can amplify concern beyond directly affected locations. Topic analysis further shows that discussions related to hydrological monitoring and traffic disruption are most strongly associated with negative sentiment, whereas rescue and coordinated-response themes align more closely with emotional recovery during later stages. Spatial regression distinguishes drivers with broadly stable nationwide effects, such as information volume and system-related topics, from drivers whose effects vary substantially across regions, including disaster exposure and population density. Compared with previous flood-emotion studies that mainly focus on temporal sentiment trends or overall online opinion patterns, this study quantitatively identifies how information intensity, system-related discourse, and local exposure conditions jointly shape spatial differences in public emotions during urban flood events. Overall, the findings demonstrate that online emotion signals can complement conventional flood-monitoring and emergency-response information by providing stage-specific and geographically differentiated support for risk communication and resilience-oriented urban flood governance.
Wang et al. (Mon,) studied this question.