Due to the numerous and complex factors that affect employee turnover, existing intelligent prediction methods are difficult to capture their dynamic changes, resulting in errors between predicted and actual values. Question: To improve the accuracy of prediction, research is conducted on an intelligent prediction method for employee turnover intention in enterprises based on recurrent neural networks. Using aspect level affective analysis to predict the emotional polarity of specific aspects of employees, extracting aspect words and analyzing their emotional polarity can effectively expand the application range of affective analysis tasks. Through RNN cycle structure and globally shared parameters, we can remember and understand complex patterns in time series data, so as to achieve accurate prediction of future data. Through visual analysis and comparison of the data of in-service employees and resigned employees, master the main influencing factors of employee turnover, and carry out descriptive analysis on the relationship between variables and employee turnover. The trained naive Bayesian classifier is used to predict turnover intention. The experimental results show that the number of resignations in this method is less than 300, which is consistent with the actual number of resignations, and gets better results. This shows that in practical application, using this prediction method to predict employee turnover is helpful for enterprises to effectively predict employee turnover in the future.