【Purpose/significance】In recent years, the haze has become more and more serious. The problem that how to study users’ emotion and the propagation characteristics of haze information information for haze events emotional early warning, providing support to emotional monitoring, and correctly guiding public sentiment is truly urgent.【 Methods/processes】This paper took the information about weibo topic"haze"of Beijing users in 2015 for this study. After data preprocessing, the article built a two-layer emotion classifier based on naive Bayes classifier algorithm to do sentiment analysis,and used social networking platform Gephi for the comments analysis.【Results/conclusions】Finally got the five groups’ :the government, the media, we media, business and the general public attitudes towards the haze and put forward the means of haze anger management and counseling, as well as the suggestions about how to correctly guide public concerns and control the network public opinions.