Abstract: 【Purpose/significance】At present, most researches on sentiment evolution in public opinion are based on topic-based meth⁃ods to analyze sentiment evolution and focus on information extracted from the text itself, and lack consideration of user characteristicsthat influence sentiment analysis in social media.【Method/process】This article fully considers user information characteristics, con⁃structs a public opinion emotion evolution method that integrates user characteristics, proposes a sentiment analysis model (U-BiL⁃STM) based on user attention mechanism, and take the COVID-19 epidemic as an example to analyze the evolution of public emotions【Result/conclusion】The research results show that the U-BiLSTM sentiment analysis model has certain advantages. The F1 value and accuracy rate can reach 97.08% and 95.19%.【Innovation/limitation】The sentiment analysis model that integrates the user attention
mechanism proposed by the research can make the analysis of the sentiment evolution of public opinion have a certain interpretability,and effectively reveal the trend of the sentiment evolution of netizens facing public health emergencies. However, due to the limitations of time and equipment conditions, only a single data source is used without considering the multi-source nature of the data in the re⁃search, and the data set for this study is not sufficient. From the perspective of research, the time dimension is only considered and the space dimension is ignored in this paper.