Abstract: 【Purpose/significance】In recent years, public emergencies have occurred frequently, and the public opinion risk caused by public emergencies has increased. How to effectively construct a network public opinion risk assessment model for public emergen⁃ cies, analyze the trend of public opinion risk, is a key link in improving the ability of public opinion risk warning and guiding network public opinion.【Method/process】Based on the disaster chain theory, this paper constructed an indicator system of public opinion risk of public emergencies from four dimensions: disaster body, disaster-affected body, disaster-resistant body, and external environment. The entropy-weight- TOPSIS method was used to evaluate the public opinion risk index of "Shenzhen Rainstorm". This research also provided a methodology for classifying public opinion warning level.【Result/conclusion】The study showed that there were significant differences in the weights of the different indicators of the disaster chain, with the disaster body having the highest weight and the disaster-resistant body the lowest. Among the first-level indicators, the heat of public opinion on the Internet and the form of public opinion dissemination had relatively higher weights, while public opinion sentiment tendency had the lowest. Among the second-level indicators, the amounts of retweets and attention paid by netizens had relatively high weights, and the geographic distribution of neti⁃ zens had the lowest weights. Meanwhile, the public opinion on "Shenzhen Rainstorm" showed that different time periods had the differ⁃ ent risk indexes, and the risk graph was similar to the actual rainfall trend in that period, which indicated that the public opinion risk index system of public emergencies constructed in this research was in line with the reality.【Innovation/limitation】The innovation of this research was to analyze the evolution law of online public opinion by combining the theory of disaster chain, and to evaluate public opinion risks hourly by adopting the entropy-weight-TOPSIS method in order to study the evolution of public opinion, and then to clas⁃ sify the level of opinion warning for public opinion monitoring and warning. However, the data source of this paper was relatively single, and only the public opinion risk is assessed, and the next step will be predictive analysis.