Abstract:In the research of emotion analysis, it is very important to use the Stacking algorithm to select the time base learners for emotion analysis. The traditional Stacking algorithm only combines different learners, does not distinguish the differences between them, and can not reflect the actual prediction of the primary learners. To solve this problem, this paper improves the Stacking algorithm based on entropy method to classify text emotion. First, the entropy method is used to determine the performance index weight of a single classifier. The weight of the index value is weighted and summed to obtain the comprehensive score of different models. The best combination of base learners is selected through the comprehensive score; Then, due to the different performance of each classifier in the base model, the prediction results after the training of the base learner are given different weights and input into the secondary learner; Finally, secondary learners are used to train and predict emotional tendencies. The experimental results show that the improved Stacking model based on entropy method is superior to the traditional Stacking model, and the selection and importance of the base learners have certain help for emotion classification, which lays a certain foundation for later text emotion analysis.