For the huge amount of comment data, the high cost of emotions is analyzed manually, and the LSTM is used to analyze the emotions of business comment. In order to train the model, including text de-duplication, removal of stop words, Chinese word segmentation, and a series of data processing, the data are coded into word vectors. Finally, the crawling data is used to verify the accuracy of the method. The result shows that the accuracy of the method is 75%.