Key points are not available for this paper at this time.
Content graphs are essential for representing domain knowledge and play a significant role in digital education. To overcome the challenges associated with manual knowledge extraction and content structuring, including low efficiency, high costs, and a high risk of errors, and to improve teaching and learning quality, this paper presents an automated approach for constructing course content graphs for digital teaching platforms by utilizing a large language model. Specifically, the pre-trained GLM-4 model is utilized for semantic parsing and entity extraction, while a group query attention mechanism is applied to infer relationships between knowledge points, automatically transforming raw course content into structured tabular datasets. The process involves two steps: entity extraction and relationship recognition, which convert the original dataset into discrete knowledge points and output them in a structured format. The structured data is then visualized using the Neo4j graph database, enabling automated content graph construction. Experimental results show that this method significantly enhances the accuracy and efficiency of content graph construction, achieving an F1 score above 0.85 for entity extraction and relationship recognition accuracies of 0.82 and 0.80 before and after processing, respectively. Leveraging the model’s strong performance, a content graph with 393,600 triples—consisting of 9 entity types and 2 relationship types—has been constructed, covering the domain of software engineering courses. These results offer an effective and scalable solution to advance the intelligent development of digital teaching platforms.
Yang et al. (Wed,) studied this question.