Towards LLM-Empowered Knowledge Tracing via LLM-Student Hierarchical Behavior Alignment in Hyperbolic Space

Xingcheng Fu; Shengpeng Wang; Yisen Gao; Xianxian Li; Chunpei Li; Qingyun Sun; Dongran Yu

cs.AIFeb 26, 2026

Towards LLM-Empowered Knowledge Tracing via LLM-Student Hierarchical Behavior Alignment in Hyperbolic Space

Authors: Xingcheng Fu, Shengpeng Wang, Yisen Gao, Xianxian Li, Chunpei Li, Qingyun Sun, Dongran Yu

Organizations: Key Lab of Education Blockchain and Intelligent Technology, Guangxi Normal University, Guilin, China · Guangxi Key Lab of Multi-source Information Mining & Security, Guangxi Normal University, Guilin, China · Computer Science and Engineering , The Hong Kong University of Science and Technology, Hong Kong, China · School of Computer Science and Engineering, Beihang University, Beijing, China

Abstract

Knowledge Tracing (KT) diagnoses students' concept mastery through continuous learning state monitoring in education.Existing methods primarily focus on studying behavioral sequences based on ID or textual information.While existing methods rely on ID-based sequences or shallow textual features, they often fail to capture (1) the hierarchical evolution of cognitive states and (2) individualized problem difficulty perception due to limited semantic modeling. Therefore, this paper proposes a Large Language Model Hyperbolic Aligned Knowledge Tracing(L-HAKT). First, the teacher agent deeply parses question semantics and explicitly constructs hierarchical dependencies of knowledge points; the student agent simulates learning behaviors to generate synthetic data. Then, contrastive learning is performed between synthetic and real data in hyperbolic space to reduce distribution differences in key features such as question difficulty and forgetting patterns. Finally, by optimizing hyperbolic curvature, we explicitly model the tree-like hierarchical structure of knowledge points, precisely characterizing differences in learning curve morphology for knowledge points at different levels. Extensive experiments on four real-world educational datasets validate the effectiveness of our Large Language Model Hyperbolic Aligned Knowledge Tracing (L-HAKT) framework.

Figures & tables

Figure 1: The bar chart labels H-all, H-ek, H-e, and H-k denote hyperbolicity measures for student-question-knowledge, question-knowledge, question-question, and knowledge-knowledge relationships, respectively. Lower values indicate stronger structural alignment with tree-like or hyperbolic space characteristics.

Figure 2: Illustration of LLM-based Teacher-student Behavior Modeling and Data Augmentation.

Figure 3: Illustration of the architecture of L-HAKT.

Model	ASSIST09		ASSIST12		EdNet		Eedi
Model	AUC(%)	ACC(%)	AUC(%)	ACC(%)	AUC(%)	ACC(%)	AUC(%)	ACC(%)
DKT	75.97 ${\pm 0.25}$	73.01 ${\pm 0.21}$	72.90 ${\pm 0.07}$	73.27 ${\pm 0.16}$	70.10 ${\pm 0.37}$	71.11 ${\pm 0.42}$	76.01 ${\pm 0.03}$	71.64 ${\pm 3.81}$
DKT+	77.21 ${\pm 0.03}$	73.32 ${\pm 0.21}$	73.64 ${\pm 0.01}$	73.42 ${\pm 0.01}$	70.20 ${\pm 0.06}$	66.21 ${\pm 1.67}$	74.32 ${\pm 0.01}$	70.79 ${\pm 0.01}$
DKT+forget	77.32 ${\pm 0.00}$	72.13 ${\pm 0.16}$	73.75 ${\pm 0.03}$	73.12 ${\pm 0.15}$	70.46 ${\pm 0.03}$	71.12 ${\pm 0.17}$	76.13 ${\pm 0.05}$	71.66 ${\pm 0.14}$
DIMKT	72.01 ${\pm 0.36}$	73.68 ${\pm 0.16}$	76.23 ${\pm 0.22}$	74.98 ${\pm 0.02}$	77.00 ${\pm 0.19}$	72.93 ${\pm 0.02}$	79.30 ${\pm 0.02}$	73.03 ${\pm 0.90}$
DKVMN	76.10 ${\pm 0.11}$	72.34 ${\pm 0.07}$	72.12 ${\pm 0.21}$	73.32 ${\pm 0.23}$	69.32 ${\pm 0.14}$	70.95 ${\pm 0.02}$	76.01 ${\pm 0.29}$	71.62 ${\pm 0.09}$

Table 1: Overall AUC and ACC performance of L-HAKT and all baselines. ACC and AUC should be as large as possible, indicating better model performance; Bold :best; Underline : runner-up.

	ASSIST09	ASSIST12	EdNet	Eedi
# Students	4160	5000	5000	5000
# Question	15643	36054	11700	26702
# Concept	167	242	1830	1050
# Interaction	206631	713123	1147423	586234

Table 2: Summary of dataset statistics.

Model	ASSIST09				ASSIST12
Model	ACC(%)	$\Delta$ (%)	AUC(%)	$\Delta$ (%)	ACC(%)	$\Delta$ (%)	AUC(%)	$\Delta$ (%)
GKT	76.32	-	72.44	-	72.39	-	73.29	-
L-HVKT(w/o hyp)	77.55	$\uparrow$ 1.61	73.78	$\uparrow$ 1.85	78.01	$\uparrow$ 7.76	76.84	$\uparrow$ 4.84
L-HVKT(w/o con)	76.98	$\uparrow$ 0.86	73.21	$\uparrow$ 1.06	76.32	$\uparrow$ 5.43	75.34	$\uparrow$ 2.80
L-HVKT	80.22	$\uparrow$ 5.11	75.10	$\uparrow$ 3.67	80.27	$\uparrow$ 10.89	77.10	$\uparrow$ 5.20

Table 3: The ACC and AUC improvements (%) results of Ablation Study. ( Bold :best; Underline :runner-up.)

Panel of Figure 4 — Figure 4: Validation of Knowledge Graph Effectiveness. (a) Compare the predictive performance of ACC and AUC in four datasets using real data, adding synthetic data, and comparing alignment.(b)Teacher Agent generates a knowledge graph by parsing rich text information.Where (.) denotes the number of related concepts.

Knowledge Tracing

Knowledge tracing models how learners’ mastery of skills changes over time from their interactions with educational tasks.

See all 31 papers

RAG-KT: Cross-platform Explainable Knowledge Tracing with Multi-view Fusion Retrieval Generation
Apr 13, 2026Zhiyi Duan, Hongyu Yuan, Rui LiuKnowledge Tracing AI in Education
Inner Mongolia University, Hohhot, China
MOSAIC: Orchestrating Collaborative Knowledge Tracing with Hierarchical Semantic Alignment
Jun 27, 2026Xinjin Li, Mengyue Wang, Yuzhen Lin +4Intelligent Tutoring Systems Hierarchical Representation Learning
Columbia University, New York, USA · University of California, Berkeley, Berkeley, USA · School of Information Systems and Management, Carnegie Mellon University, Pittsburgh, PA, USA +4
ConceptKT: A Benchmark for Concept-Level Deficiency Prediction in Knowledge Tracing
Mar 25, 2026Yu-Chen Kang, Yu-Chien Tang, An-Zi YenKnowledge Tracing Intelligent Tutoring Systems
Department of Computer Science, National Yang Ming Chiao Tung University, Taiwan

Apr 13, 2026J/K move · Enter open · S save

RAG-KT: Cross-platform Explainable Knowledge Tracing with Multi-view Fusion Retrieval Generation

Zhiyi Duan, Hongyu Yuan, Rui Liu

Inner Mongolia University, Hohhot, China

Knowledge Tracing AI in Education Retrieval-Augmented Generation

Open paper AskPDF arXiv

Knowledge Tracing (KT) infers a student's knowledge state from past interactions to predict future performance. Conventional Deep Learning (DL)-based KT models are typically tied to platform-specific identifiers and latent representations, making them hard to transfer and interpret. Large Language Model (LLM)-based methods can be either ungrounded under prompting or overly domain-dependent under fine-tuning. In addition, most existing KT methods are developed and evaluated under a same-distribution assumption. In real deployments, educational data often arise from heterogeneous platforms with substantial distribution shift, which often degrades generalization. To this end, we propose RAG-KT, a retrieval-augmented paradigm that frames cross-platform KT as reliable context constrained inference with LLMs. It builds a unified multi-source structured context with cross-source alignment via Question Group abstractions and retrieves complementary rich and reliable context for each prediction, enabling grounded prediction and interpretable diagnosis. Experiments on three public KT benchmarks demonstrate consistent gains in accuracy and robustness, including strong performance under cross-platform conditions.

Abstract

Knowledge Tracing

Explore similar work

RAG-KT: Cross-platform Explainable Knowledge Tracing with Multi-view Fusion Retrieval Generation

MOSAIC: Orchestrating Collaborative Knowledge Tracing with Hierarchical Semantic Alignment

ConceptKT: A Benchmark for Concept-Level Deficiency Prediction in Knowledge Tracing