With the development of new social media, Weibo, as an important media for the dissemination of public opinion, has become a platform for the excavation of public opinion. Natural Language Processing technology can extract effective emotional information from Weibo texts, and provide scientific decision-making basis for monitoring network public opinion, forecasting potential problems, and product analysis. In order to overcome the limitation of the existing shallow learning algorithm for complex function expression, this study attempts to integrate the idea of deep learning, and puts forward an improved recurrent neural network based on Word2Vec and long-term memory network to analyze Chinese Weibo emotion. In the more than 20 000 Chinese corpus of training experiment, the experimental data with SVM, RNN, and CNN are compared, comparison results show that the emotion analysis model proposed in this study reaches the accuracy rate of 91.96%, thus it can effectively improve the accuracy of the Weibo text sentiment classification.