The emotion of online business review texts was mined, and different features of review texts were fused to provide more information for the classifier. A new online e-commerce emotion classification algorithm was proposed. Firstly, aiming at the deficiency that the the emotional information features of words cannot be fused well in the traditional word embedding model, the multi-feature fusion method of word embedding features and part-of-speech features was considered in this study. Secondly, on the basis of two feature fusion methods, the classification accuracy based on both two-channel and single-channel methods was compared. A parallel CNN and BiLSTM-attention two-channel neural network model was proposed. Finally, the real JD.COM e-commerce review data set was used to evaluate the proposed model, and compared with different classification algorithms in experiments. Experimental results show that the new hybrid method has better classification accuracy, recall rate, and F1 scores.