GLLM-KT: A Graph-Incorporated Ultra-small Large Language Model for Knowledge Tracing

Knowledge Tracing (KT) aims to model students’ latent knowledge states from historical learning interactions. However, existing KT methods often struggle with generalization across diverse datasets. Furthermore, while Large Language Models (LLMs) have shown potential for improving KT, their enormous parameter sizes limits practical deployment. To address these issues, we propose GLLM-KT, a graph-incorporated ultra-small LLM-based KT model. It integrates a knowledge prerequisite graph to capture concept dependencies, a structured instruction template for task formulation, and an adaptive confidence head to improve prediction robustness. In this paper, we investigate the potential of ultra-small large language models for KT. Extensive experiments on four benchmark datasets demonstrate the superior performance of GLLM-KT in prediction accuracy and generalization across different datasets and scales of ultra-small LLMs.