The public opinion of recurrent events is the existence of normalization in the network society. The targeted research and scientific prediction of it will help to guide the orderly development of public opinion and maintain the stability of the network society. This study constructs a risk assessment index system of network public opinion for recurrent events, uses Delphi method and analytic hierarchy process to test and measure the weight of the index system. Based on this, taking the "Garbage Classification Event" as an example, this study analyzes the characteristics of indicators such as public opinion subject, public opinion object and key communication nodes in each stage of public opinion through statistical analysis, principal component analysis and text emotion analysis, especially compares the similarities and differences between the recurrent stage and other stages. The results provide a reference for the effective identification, risk assessment and scientific guidance of recurrent public opinion.