[Purpose/Significance] Large-scale research infrastructures represent a symbol of national scientific competitiveness, and the scientific data they generate play a crucial role in scientific research, technological advancement, and economic development. However, traditional data management methods face challenges such as data redundancy, difficulties in sharing, and repetitive experiments, urgently requiring empowerment through next-generation artificial intelligence technologies. This paper aims to study a method for constructing an AI-enabled scientific data management and recommendation framework for large-scale research infrastructures, focusing on exploring a personalized recommendation system architecture that integrates knowledge graphs and large language models to enhance the integration and intelligent recommendation capabilities of scientific data. [Method/Process] This paper proposed an artificial intelligence-based framework for scientific data management and recommendation for large-scale research infrastructures. It introduced technologies such as knowledge graphs, link prediction, and large language models (LLMs), proposing a personalized data recommendation system architecture oriented towards the entire research process. [Result/Conclusion] The system architecture includes modules for data collection, knowledge extraction, semantic reasoning, and personalized recommendation. The data layer is responsible for the organization and integration of multi-source heterogeneous data, the analysis layer performs in-depth analysis using graph neural networks and large language models, and the application layer provides accurate recommendation services to researchers. The framework proposed in this paper contributes to improving data utilization, promoting scientific and technological innovation capabilities, and enhancing the integration and intelligent recommendation of multi-source heterogeneous data.