With the rapid development of the Internet, the amount of data generated by users is exploding. Accurately mining user interests from massive data and realizing personalized recommendation has become an important research topic in the field of information services. Personalized recommendation systems can enhance user experience and bring significant economic benefits to businesses. The aim of this study is to construct a personalized Artificial Intelligence (abbreviated to AI) recommendation system based on big data science, which integrates multi-source heterogeneous data to achieve deep analysis of user behavior and intelligent recommendation, which can improve the accuracy and real-time performance of recommendations. This article studies the use of MapReduce distributed computing framework for preprocessing massive data, combined with collaborative filtering, content recommendation, and real-time recommendation algorithms to design and implement an integrated recommendation platform. By introducing user interest models and dynamic update mechanisms, the system enhances its responsiveness to changes in user behavior. Through extensive experimental verification, real-time recommendation algorithms are superior to traditional recommendation algorithms in key indicators such as accuracy, recall, and coverage. The experimental results show that real-time recommendation algorithms can improve the platform’s commercial conversion rate. The innovation of this study lies in proposing a personalized recommendation architecture that combines big data distributed processing with multi algorithm fusion, effectively improving the scalability and real-time response capability of the system.