Organizations: State Key Laboratory of Cognitive Intelligence, University of Science and Technology of China, Hefei, China · iFLYTEK AI Research, Hefei, China · State Key Laboratory of Cognitive Intelligence, University of Science and Technology of China & Institute of Artificial Intelligence, Hefei Comprehensive National Science Center, Hefei, China
Knowledge Tracing (KT) aims to assess students' dynamic knowledge states from their learning histories. While most existing KT methods focus on single-domain learning with notable success, real-world learning scenarios often involve multiple domains simultaneously, introducing two critical factors: 1) Cognitive load, arising from managing learning across domains in both temporal and knowledge dimensions. 2) Knowledge transfer, where knowledge states in one domain influence related states both within and across domains. In this paper, we focus on exploring these factors to improve students' knowledge state assessment in multi-domain learning scenarios and propose a novel method incorporating cognitive Load and knowledge Transfer for Multi-domain Knowledge Tracing (LT-MKT). Specifically, to bridge isolated domains, LT-MKT first integrates textual information from questions and their associated concepts to construct a Multi-domain Hierarchical Graph, leveraging the advanced representational capabilities of large language models (LLMs). Then, cross-domain features in both the temporal and knowledge dimensions are explicitly modeled to capture the effects of cognitive load. Additionally, a knowledge transfer module is designed to model the propagation of knowledge states within and across domains. By jointly modeling these factors, LT-MKT enables more accurate prediction of students' future performance. Finally, extensive experiments on real-world datasets demonstrate that our method achieves state-of-the-art performance.
Figures & tables
Figure 1. Illustration of Multi-domain Knowledge Tracing. (a) Hierarchical Question Concept Graph outlines relationships between concepts, as well as between questions and concepts. Questions in one domain (e.g., q3 ) may link to concepts from another (e.g., c2 ), and concepts across domains can also correlate (e.g., c2 and c3 ). (b) In multi-domain learning scenarios, students may alternate between domains like math, physics, and chemistry, enhancing learning through knowledge transfer but also increasing cognitive load.
Figure 2. Overall framework of LT-MKT. The model first constructs a multi-domain hierarchical graph from question texts and knowledge concepts with LLM-guided reasoning. Based on this graph, LT-MKT models cognitive load through question difficulty, domain transition, and domain coverage, and updates students’ knowledge states via GRU-based state evolution. It then performs domain-aware knowledge transfer with intra-domain prerequisite propagation and inter-domain correlation propagation, followed by state fusion and performance prediction for the next interaction.
Datasets
JuniorH
SeniorH
PTADiscJP
PTADiscDS
#Students
1,081
4,869
29,430
12,271
#Concepts
139
198
1,245
634
#Questions
170
264
27,820
18,702
#Interactions
39,230
133,683
11,172,165
1,788,245
#Avg.Inter
36.29
27.45
379.6
145.73
#Avg.Cross
14.26
12.39
8.72
7.81
Table 1. Dataset statistics
Methods
JuniorH
SeniorH
PTADiscJP
PTADiscDS
AUC ↑
ACC ↑
RMSE ↓
AUC ↑
ACC ↑
RMSE ↓
AUC ↑
ACC ↑
RMSE ↓
AUC ↑
ACC ↑
RMSE ↓
DKT
0.8840
0.7991
0.1338
0.8861
0.7926
0.1377
0.7246
0.7843
0.3687
0.6504
0.7602
0.4358
GKT
0.8818
0.7983
0.1324
0.8883
0.7828
0.1391
0.7298
0.7892
0.3664
0.6534
0.7646
0.4335
AKT
0.8871
0.8032
0.1341
0.8864
0.7928
0.1382
0.7307
0.7904
0.3655
0.6543
0.7651
0.4337
HawkesKT
0.7515
0.7202
0.2273
0.7610
0.7124
0.2376
0.7441
0.7903
0.3612
0.6574
0.7641
0.4358
LPKT
0.8844
0.8004
0.1335
0.8850
0.7922
0.1383
0.7274
0.7876
0.3669
0.6575
0.7634
0.4346
Table 2. Results of all comparison methods on the student performance prediction task. Existing state-of-the-art results are marked by the underline, and the best results are bold. * indicates p-value <0.05 in the t-test.
Methods
JuniorH
SeniorH
PTADiscJP
PTADiscDS
AUC ↑
ACC ↑
RMSE ↓
AUC ↑
ACC ↑
RMSE ↓
AUC ↑
ACC ↑
RMSE ↓
AUC ↑
ACC ↑
RMSE ↓
w/o CL
0.9005
0.8116
0.1322
0.8935
0.8150
0.1268
0.7471
0.8237
0.3610
0.6673
0.7896
0.4231
w/o TG
0.9227
0.8322
0.1219
0.9160
0.8360
0.1250
0.7594
0.8302
0.3544
0.6799
0.7955
0.4102
w/o SF
0.9320
0.8419
0.1227
0.9252
0.8455
0.1152
0.7603
0.8388
0.3505
0.6865
0.8011
0.4097
LT-MKT
0.9387
0.8425
0.1214
0.9312
0.8470
0.1141
0.7645
0.8410
0.3485
0.6929
0.8031
0.4065
Table 3. Results of ablation experiments across four datasets.
LLM Backbone
ER ↑
SC ↑
Fleiss’ κ↑
Qwen2.5-7B
3.72
3.60
0.61
Qwen2.5-32B
4.07
3.95
0.69
Qwen2.5-72B
4.29
4.16
0.74
Llama3-70B
4.11
4.02
0.71
DeepSeek-R1
4.13
3.98
0.68
Claude3 Sonnet
4.45
4.38
0.79
Table 4. Expert evaluation results of graph quality generated by different LLM backbones.
Figure 3. Analysis of cognitive-load representations.
Figure 4. Results in cold start scenarios.
Figure 5. A case study of a student’s cross-domain knowledge state evolution under LT-MKT.
Figure 6. Parameter sensitivity analysis of λP and ws .
Knowledge tracing (KT) aims to predict students' future performance by modeling their evolving knowledge states from historical interactions. Existing KT methods usually treat the raw interaction sequence as a unified behavioral process, overlooking the phase-specific nature of learning behaviors. Our preliminary observations show that students are more likely to correctly answer previously failed knowledge concepts after sufficient practice, suggesting a transition from ability-building to proficiency-oriented learning. Motivated by this, we propose Phase-Aware Knowledge Tracing (PAKT), a KT framework that decomposes student interactions into ability and proficiency phases based on the tailored decomposition mechanism. To effectively exploit the decomposed sequences, we design a multi-branch Transformer with a type-aware readout module to jointly capture phase-specific and holistic knowledge states. We further provide a causal analysis to reveal the confounding bias caused by entangling complex learning behaviors in phase-agnostic KT models. Extensive experiments on six public benchmarks demonstrate that our method consistently outperforms representative baselines, with a maximum AUC gain of 1.33% and an average gain of 0.82%.
Duantengchuan Li, Yingqian Bi, Jinsong Chen +2
School of the Information Management, Wuhan University, Wuhan 430072, China · School of Computer Science and Artificial Intelligence, Hubei University of Technology, Wuhan 430068, China · Faculty of Artificial Intelligence in Education, Central China Normal University, Wuhan 430074, China
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.
Xingcheng Fu, Shengpeng Wang, Yisen Gao +4
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 +1
Knowledge Tracing (KT) models students' knowledge states based on learning interactions to predict performance. While deep learning-based KT models have boosted predictive accuracy, most models rely on deterministic vector embeddings and opaque latent state transitions, limiting interpretability regarding how specific past behaviors influence predictions. To address this limitation, we propose Probabilistic Logical Knowledge Tracing (PLKT), an interpretable KT framework that formulates prediction as a goal-conditioned evidence reasoning process over historical learning behaviors. Instead of representing knowledge states as deterministic vector embeddings, PLKT employs robust Beta-distributed probabilistic embeddings to represent student knowledge states. This probabilistic foundation allows us to model the uncertainty of historical behaviors and perform explicit logical operations (e.g., conjunction), constructing transparent reasoning paths that reveal how specific past interactions contribute to the prediction. Extensive experiments show that PLKT outperforms state-of-the-art KT methods while achieving superior interpretability. Our code is available at https://anonymous.4open.science/r/PLKT-D3CE/.
Siyu Wu, Cong Xu, Wei Zhang
1Shanghai Institute of AI Education, East China Normal University, Shanghai, 200241, China · 2Shool of Computer Sicence and Technology, East China Normal University, Shanghai, 200241, China