This study proposes a computerization intelligent retrieval and personalized and accurate push algorithm based on reinforcement learning, aiming to improve the accuracy of information retrieval and the personalization of push. By deeply analyzing user behavior data and preference information, an adaptive dynamic environment model is constructed to realize personalized information push for different user needs. The reinforcement learning strategy is introduced to continuously optimize the push content through the feedback mechanism, so that the push results gradually fit the user's interest preferences, thereby improving the user experience. In addition, the use of computerization processing methods significantly improves the system's computing efficiency and real-time response capabilities, and effectively reduces the delay problem in the process of information retrieval and push. The experiment uses a test set containing a large amount of user behavior data, and selects several key evaluation indicators, including information retrieval accuracy, push personalization, computing efficiency and user satisfaction. The experimental results show that the algorithm has improved the accuracy of information retrieval by more than 20% compared with traditional methods, and the matching degree of push content and user preferences has been significantly improved. In addition, the user satisfaction score has increased by 15% compared with the traditional method, indicating that users are more recognized for the relevance and personalized expression of the push content. At the same time, the real-time response capability of the system has been improved by 30%, ensuring rapid responsiveness in a multi-user environment.