Dynamic monitoring and early warning of public emotional perception of risk during extreme rainstorm disasters: A study based on social media

For effective management of extreme rainstorm disasters, a timely understanding of the public emotional perception of risk is crucial to enhancing governmental risk communication and emergency response strategies. Therefore, this study develops a sentiment analysis framework that combines a rainstorm-specific sentiment lexicon with a deep learning model. By utilizing large-scale social media data, the framework further achieves dynamic monitoring and early warning of the public emotional perception of risk during extreme rainstorm events. Specifically, this paper employed text mining techniques to analyze the emotional features of 51,222 Weibo posts related to rainstorm disasters, thereby constructing a specialized sentiment lexicon. The lexicon was then integrated into a TextCNN model to create a knowledge-enhanced hybrid sentiment analysis model. Compared with baseline models including GPT-4o, LLaMA-3, and RoBERTa, this hybrid method achieved optimal performance across all evaluation metrics, with accuracy and F1-score improvements of 10.9% and 9.9%, respectively. Moreover, an empirical analysis of the 2023 Zhuozhou extreme rainstorm disaster validated the framework's efficacy. Findings reveal that our method can effectively monitor the public emotional perception of risk and provide early warning of risk anomalies during severe rainstorms by using the emotion index, which yields valuable insights for governmental bodies to accurately understand public risk perception and the dynamic evolution of disaster scenarios.