LLM-Driven Ontology Learning to Augment Student Performance Analysis in Higher Education

In educational settings, a challenge is the lack of linked and labeled data, hindering effective analysis. The integration of ontology facilitates the formulation of educational knowledge concepts, student behaviors, and their relations. Traditional ontology creation requires deep domain knowledge and significant manual effort. However, advancements in Large Language Models (LLMs) have offered a novel opportunity to automate and refine this process. In this paper, we propose an LLMs-driven educational ontology learning approach aimed to enhance student performance predictions. We leverage LLMs to process lecture slide texts to identify knowledge concepts and their interrelations, while question texts are used to associate them with the concepts they assess. This process facilitates the generation of the educational ontology that links knowledge concepts and maps to student interactions. Additionally, we deploy a dual-branch Graph Neural Network (GNN) with distance-weighted pooling to analyze both global and local graph information for student performance prediction. Our empirical results demonstrate the effectiveness of using LLMs for ontology-based enhancements in educational settings.