An emotion-based independent cascade model is proposed to analyze the process of sentiment spreading. User features, structural features and tweet features are introduced into the learning model for finding the sentiment changes of retweets. The historical information based transforming weights are proposed for the sentiment prediction of retweets.