[Purpose/significance] The sentiment classification is still one of the cross-cutting issues needed to focused on.[Method/process] With the help of emotion unrelated words, by the spectral clustering algorithm, the authors constructed a cross-domain feature words emotion cluster in the source and target areas of the field. The position of the features and characteristics of emotional words, keyword features, and POS features were integrated into the logic of the regression classification algorithm to achieve a cross-cutting emotion classification algorithm based on the multi-feature fusion. [Result/conclusion] Research results show that CDFF (Cross-domain pulse Four Factors) algorithm is effective when the cross-domain user emotion is classified and its provide reference for same study.