Opinion leaders can disseminate information,expand influence and guide public opinion in network events. To identify opinion leaders,this paper introduces multilevel text sentiment analysis to classify texts into "very positive", "positive","neutral","negative",and"very negative",which can detect the support degree of netizens for bloggers more precisely. On this basis,the paper further fuses multiple user features to construct a systematic framework of identifying online opinion leaders,and uses the data from Sina Weibo for experimental study. The effectiveness and reliability of the proposed method is proved by being compared with the models without the index of support degree or without two level support degrees.