Recommendation systems can analyze users' historical behavioral data, discover potential patterns from massive raw data, and push personalized learning resources that may be of interest to learners based on their behavioral characteristics, alleviating the problem of information overload. Traditional collaborative filtering algorithms and other mature recommendation technologies limit the performance of recommendation algorithms when user historical data is sparse. Knowledge graph has a wide range of applications in natural language processing, intelligent question answering, information retrieval, recommendation systems, and intelligent dialogues. It can be embedded into a continuous vector space to solve data sparsity and cold start problems. This article conducts research based on knowledge graph technology to provide a complete solution for personalized recommendation systems for learning resources. The main contributions include: firstly, researching the construction of knowledge graphs, integrating entities and relationships to form a structured knowledge graph; Secondly, research the construction of mathematical models to achieve deep mining of structured and unstructured data; The third is the system architecture design, where each layer is independent of each other, making it easier to modify and expand each layer.