With the rapid development of the times, human resource management systems are gradually unable to keep up with the current development speed. Most current human resource management systems are designed from a recruitment perspective, lack consideration of prior intervention in personnel flow, and have poor role adaptability. This paper used J2EE to design a human resource management system, developed web and mobile terminals, and introduced the LSTM (Long Short-Term Memory) model to predict human resource loss in advance. The study first used J2EE technology to develop the Web, and used the Android platform to develop the mobile terminal, and then used the RESTful mode and WebService interaction technology to communicate between the front and back ends, and the Android platform and the Web terminal. Finally, the LSTM model was introduced to build a human resource loss prediction model, which was applied to the system to predict the loss of company employees, and RabbitMQ technology was used to realize the interaction between the employee loss prediction model and the system. The experimental results show that the prediction accuracy of the human resource management system developed by J2EE-LSTM reaches 0.96, and the response time is only 4.5s under 1,300 concurrent users. The experiment shows that the introduction of J2EE-LSTM to develop a human resource management system can significantly improve user satisfaction and the accuracy of human resource loss prediction, providing an important reference for enterprises to retain talents.