LDA thematic model and LSTM model were constructed to solve the problem of emotional classification of short text. The feature representation method of comment information is extended by word embedding method and LDA model to solve the problems of sparse and inconspicuous features in the short text. We use LSTM to build a classification model,which can memorize long-term in-formation and better learn text characteristics by constantly learning text context information. A comparative experiment was conducted on the evaluation dataset. Compared with the traditional classification model,the emoti...