Abstract: This paper aims to reveal the relationship between topics of micro-blog and user behavior at different stages of major emergency. The LDA model improved by the relevance formula is used to extract the topic of micro-blog, and cosine similarity algorithm is used to calculate the correlation between micro-blog information topic and user behavior, so as to analyze the behavior of micro-blog users under different topics during the epidemic. During the major emergency, the evolution trend of users' retweets, comments and favorites was similar. Among them, the Spearman correlation coefficients of them are all above 0.4 and have significant correlations. In addition, micro-blogs related to the topic of pronouncing opinions and information disclosure are likely to be favorite. Micro-blogs related to the topic of pronouncing opinions, progress of the events and information disclosure are likely to be commented on. During the major emergency,the user's behavior shows obvious tendentiousness,which will produce different behavior for different types of topics.

Key words: Micro-blog topics, User behavior, Major emergency, Public opinion evolution, LDA model, Public opinion management