Traditional long-short term memory network (LSTM) cannot automatically select the most important latent semantic factors in text categorization. To solve the problem, this paper proposes an improved LSTM model. First, the traditional LSTM operation relationship is extended to the bidirectional mode, so that the network fully remembers the context of the input feature words. Then, the pooling layer is added in front of the output layer to better select the most important latent semantic factors. The experiment on the Internet Movie Database review data show that the model is superior to the traditional long-short term memory neural network and other similar models, revealing that the improved scheme proposed in this paper can improve the accuracy of text classification.