LLM4EduKG: LLM for Automatic Construction of Educational Knowledge Graph

The field of education is undergoing a significant transformation towards digital and intelligent education, driven by advancements in artificial intelligence. Knowledge graphs (KGs), as a structured representation of knowledge and information, offering a powerful way to integrate diverse and multi-sourced heterogeneous data from across the Internet. The current methodologies for constructing educational knowledge graphs, however, are confronted with challenges including labor-intensive, time-consuming, and the necessity for substantial computational resources, which severely limit their practical application, especially in resource-constrained environments. In this paper, we proposed an LLM-based automatic construction method to alleviate the labor and time consumption in existing methods, and further explored LLM’s capabilities in Chinese-speaking context. Specifically, we designed a structured prompt framework to automatically extract and evaluate educational triples generated from original text. The prompt encompasses both task and model dimensions, allowing for flexible adjustments to different tasks and models, thus significantly improved the transferability of our method. Comparative experimental results from two real-world Chinese-datasets, across four advanced LLMs, demonstrate the effectiveness of the proposed method. We believe that our work represents a significant attempt by the LLM in the field of education.