LLM-Powered Construction of Course Knowledge-Competency Graphs

With the rapid growth of knowledge and the rising demand for personalized education, intelligent digital technologies have become fundamental to course instruction, among which knowledge graph technologies are gaining increasing traction. Addressing the limitations of traditional knowledge graph construction methods, such as heavy reliance on manual annotation, low efficiency, poor generalization, and the lack of competency-level modeling, this paper proposes a large language model (LLM)-enhanced, three-dimensional “Knowledge–Competency” graph construction approach. Firstly, leveraging LLMs combined with prompt engineering, course knowledge points and their relationships are efficiently extracted, enabling the automated construction of course knowledge graphs. Then, a general domain competency graph is adapted to the course context to generate a course-specific competency dimension graph. This graph is further integrated with the knowledge graph through rule-based fusion, forming a practical instructional application layer. Experiments verify the effectiveness of the proposed method in knowledge extraction tasks, especially in relation extraction where the accuracy is improved by 18.75% and the F1 score by 16.99% comparedwith deep learning methods. Meanwhile, the feasibility and potential of large language models (LLMs) in constructing course graphs under low-resource conditions are demonstrated. The multi-level graph structure not only extends the expressive capacity of traditional knowledge graphs but also provides structured support for learning outcome evaluation, path planning, and personalized recommendation.