Text emotion analysis based on deep learning is an important direction of natural language processing research at present. Based on CNN and Bi-LSTM this paper proposes a emotion analysis algorithm with better performance than the previous two algorithms.The improved emotion analysis algorithm combines the traditional deep learning structure and combines the Bi-LSTM,CNN and Attention mechanism. The part combining the Bi-LSTM and Attention mechanism is mainly used to extract global features and focus on target words.The part of CNN and Attention mechanism is mainly used to extract local important features.Finally, the characteristics of the two parts are integrated and then classified. The results of the experiment showed that when analyzing the emotions of online comments , The F1 of CNN model was 0.7939, the F1 of BI-LSTM was 0.7959, the F1 of BI-LSTM-Attention was 0.7998,and the F1 of BI-LSTMA-CNNA was 0.8028.Therefore, the performance of the improved model is better than the other three models.