Research on Knowledge Graph Extraction Methods for Chinese STEM Curriculum

STEM education, as an innovative teaching model, has gained widespread attention in recent years. However, the lack of relevant textbooks and learning resources has made its implementation challenging. Developing interdisciplinary knowledge graphs tailored for STEM education has become an urgent issue. To address this, a knowledge extraction framework named Llms4edu is proposed, which utilizes a series of effective prompts to guide large language models in knowledge extraction. Specifically, the knowledge extraction task is transformed into multiple rounds of question-and-answer interactions with the LLM, gradually identifying entity-relation triplets from subject data. Through experiments, an F1-score of 89.4% was achieved on the named entity recognition task in the chemistry subject, and an F1-score of 66.7^% on the relation extraction task. Finally, a subject ontology model was built for subject text, and a subject data set was constructed using Llms4edu, which includes three subjects of junior high school mathematics, physics, and chemistry, a total of 2,511 entities, 2,010 relationship triples, and cross-disciplinary knowledge is linked to construct a cross-disciplinary knowledge graph.