Abstract: 【Purpose/significance】To explore the evolutionary patterns of user emotion in unexpected public opinion events in cyber⁃ space and their underlying influencing factors, aiming to provide reference for the practical efforts in network governance.【Method/ process】Based on users' emotion, this paper extracts and cleans Weibo users' comments during the "Shanghai Epidemic" period, and realizes sentiment classification of comments by using CNN-BiLSTM network model. Based on the results of sentiment analysis, the Latent Dirichlet Allocation (LDA) model and the qualitative coding are used to realize the recognition of positive and negative emo⁃ tional topics and the extraction of emotional evolution influencing factors, which help explore the distribution characteristics of emo⁃ tional evolution influencing factors and the correlation with the scale of public opinion and user emotion.【Result/conclusion】The study found that the five emotional evolution factors of event, subject, regulation, emergency and time show different characteristics in the evolution process, and affect the scale of public opinion. As the number of factors increases, the user discussion period becomes more complex, while the users’emotional fluctuations also become more frequent and intense. Based on these results, this paper puts forward the corresponding public opinion governance suggestions.【Innovation/limitation】Based on the CNN-BiLSTM model and the "time - scale - emotion" three - dimensional evolution model, the research innovatively explores the influencing factors of emotional changes in sudden public opinion. In the future, different events and more platforms' review data can be combined to further enhance the granularity and universality of research.