Abstract: AI application developers often need to learn various deep learning models during the development process, but it is not easy to learn deep learning models from the huge amount of data because the AI field lacks a unified platform for learning deep learning models, and AI application developers are not necessarily experts in the AI field. Therefore, a knowledge graph construction method of deep learning model is proposed. The method extracted a variety of knowledge, such as model and implementation, components and implementation, component dependencies, high-level concepts, component categories, component characteristics, component descriptions, and component open relationships, to construct a model knowledge graph. The experimental results show that the knowledge graph is of high quality with a tuple correct rate of 90.63%, which can provide a platform for AI application developers to learn deep learning models and improve development efficiency.