In 2008, ZHANG Song-Mao, a researcher of Chinese Academy of Sciences, proposed the application of 3D animation automatic generation technology for mobile phone SMS. The sentiment analysis of SMS is an important part of the 3D animation automatic generation system. At present, the method used in the system is a traditional machine learning method, which has a low accuracy and cannot achieve a practical purpose. In recent years, deep learning has achieved good results in the task of sentiment analysis. Convolutional neural network can automatically extract the semantic and sentiment features of text messages, and attention mechanism can automatically obtain weighting information for words. Therefore, this study proposes to apply the attention mechanism and convolutional neural network in deep learning to the classification of sentiment analysis in the system of SMS automatic generation. Experiments show that the convolutional neural network based on attention mechanisms has significantly improved the accuracy, recall rate, and F-value than the previous methods.