Abstract: In this paper we present a novel approach for long-term prediction of the development trend of Web sentiment. For each class of social events, the class model library of the development trend of Web sentiment is established by cycle analysis and hierarchical clustering. Then the adaptive transform is applied to the already known development trend of a new social event, and the min-sum of MSE from the library is selected to predict the future development trend of web sentiment. Experiments show that, compared with the traditional methods, the approach presented in this paper yields a higher correlation in predicting the long-term development trend of web sentiment, and can predict the turning points of the development trend more effectively.