In the realm of digital education, Educational Recommender Systems (ERSs) play a pivotal role in enhancing learning outcomes. Nevertheless, existing systems encounter significant challenges, including data sparsity, cold-start problems, limited dynamic adaptability, insufficient personalization, and poor interpretability. To tackle these issues, we introduce an innovative personalized recommendation framework that integrates knowledge graphs (KGs) with Large Language Models (LLMs). By utilizing LLMs to enrich knowledge graphs—such as through the automatic generation of resource tags, difficulty levels, and overviews—our framework effectively addresses data sparsity and cold-start problems. It dynamically updates the knowledge graph and recommendation strategies based on real-time changes in students’ knowledge states, while also enabling learners to adjust their knowledge profiles via Open Learner Models (OLMs). Our recommendation module synergizes knowledge graphs with advanced algorithms to compute the similarity between students and educational resources, facilitating personalized ranking. To enhance the interpretability of recommendations, our system extracts relevant subgraphs and articulates the recommendation rationale through natural language generation techniques. Empirical evidence demonstrates that this framework not only significantly boosts the quality and efficacy of recommendations but also fosters personalized education and elevates the intelligence of educational systems.