The recommendation system can help users automatically select items of interest from a large amount of information. The use of artificial intelligence technology to find out the willingness to express emotions in the text has received widespread attention, but data sparseness and cold start problems have seriously affected the quality of recommendations. Based on this, we propose a Text-based Attention Neural Network (TANN) recommendation model, which is based on text perception. First, use word2vec to obtain the word vector representation of the user comment text. Then, the combination of LSTM and attention mechanism is used to capture the contextual information in the text, and finally the emotional factors are classified and predicted through the CNN layer. Compared with the advanced baseline method, the TANN recommendation model is better than other baselines and can effectively solve the user's data sparseness and cold start problem.