In order to explore the characteristics and laws of the evolution of network public opinion, it is necessary to effectively identify high-value public opinion topics from a large number of data of network public opinion at the content level, and integrate the public opinion content isolated at different time points with the time dimension. This paper combines information science theory, life cycle theory, public opinion communication theory, social network analysis method and text analysis method, proposes the research design based on community detection and key node identification. Finally, the paper takes the "Shanghai stampede" incident of Sina Weibo as the research object for empirical analysis. The results show that: the addition of non-text feature elements such as user attributes and user behavior to the topic discovery makes up for the lack of user relationship and improves the efficiency of topic discovery; the proposed topic discovery method reduces the impact of sparsity of micro-blog text; the research finds out the changing state of the subject content of public opinion events in the whole life cycle. The proposed research design can provide effective methodological support for relevant decision-making, and the research conclusion has information reference value.