HGC-FKT: 异构反事实图增强的公平知识追踪模型

doi:10.3969/j.issn.1000-5641.2026.05.007

摘要/Abstract

摘要：

为解决现有知识追踪模型存在的群体预测性能差异、数据稀疏性及遗忘规律建模不足3个核心问题, 提出了一种异构反事实图增强的公平知识追踪模型HGC-FKT (Heterogeneous Counterfactual Graph-Enhanced Fair Knowledge Tracing). 为缓解不同学生群体之间的预测性能差异, 本文设计了一种反事实启发的图增强机制. 该机制根据历史学习表现识别弱势学生群体, 并从其未作答问题中筛选高价值候选问题, 构造“若该学生与候选问题发生交互”的假设性学习情境; 进一步利用相似学习者的已观测行为对潜在作答结果进行经验估计, 据此扩充弱势学生的图连接, 并结合群体差异正则化降低不同学生群体之间的预测性能差距. 同时, 构建学生–问题图和问题–知识概念图的双图结构, 融合行为与结构信息, 并融入遗忘特征以优化建模. 在Assist09、Assist12、Junyi数据集上的实验表明, HGC-FKT在预测公平性上优于现有主流方法, 结合图结构建模与遗忘感知注意力机制, 模型提升了建模精度与认知贴合度; 在Assist09数据集上, 模型的AUC (Area Under the ROC Curve) 值达到0.8108.

关键词: 知识追踪, 反事实图增强, 预测公平性, 异构图卷积网络

Abstract:

Knowledge tracing (KT) aims to infer learners’ dynamic knowledge states from historical interactions and predict their future responses. Despite considerable advances, existing KT methods still suffer from three practical limitations: performance disparities across student groups, sparse interaction data, and insufficient modeling of forgetting dynamics. To address these challenges, this paper proposes HGC-FKT, a heterogeneous counterfactual graph-enhanced fair knowledge tracing model. The model introduces a counterfactual-inspired graph augmentation module, which identifies disadvantaged student groups based on their historical learning records and constructs hypothetical interactions with high-value candidate questions. Potential responses to these unobserved interactions are estimated from the observed behaviors of similar learners and served as augmentation signals to enrich the student-question graph. Meanwhile, a group-disparity regularizer further reduces prediction differences between student groups. To alleviate interaction sparsity, HGC-FKT jointly encodes a student-question graph and a question-knowledge concept graph, thereby integrating behavioral and structural information. Furthermore, a forgetting-aware attention module further incorporates actual time intervals and repeated practice into temporal knowledge-state modeling. Extensive experiments on the Assist09, Assist12, and Junyi datasets demonstrate that HGC-FKT consistently improves predictive performance for disadvantaged students while maintaining strong overall accuracy. The model achieves an AUC of 0.8108 on Assist09, demonstrating the effectiveness of integrating graph augmentation, dual-graph representation learning, and forgetting-aware temporal modeling designs.

Key words: knowledge tracing, counterfactual graph augmentation, predictive fairness, heterogeneous graph convolutional network

中图分类号:

TP183

于洋, 王雅婷, 张天成, 于明鹤, 于戈. HGC-FKT: 异构反事实图增强的公平知识追踪模型[J]. 华东师范大学学报（自然科学版）, 2026, 2026(5): 77-94.

Yang YU, Yating WANG, Tiancheng ZHANG, Minghe YU, Ge YU. HGC-FKT: Heterogeneous counterfactual graph-enhanced fair knowledge tracing model[J]. J* E* C* N* U* N* S*, 2026, 2026(5): 77-94.

图/表 11

图1

图2

表1

表2

表3

表4

表5

表6

图3

图4

图5

参考文献 27

1	Liu Z H, Liu S N, Gu W F.. Graph knowledge structure for attentional knowledge tracing with self-supervised learning. IEEE Access, 2025, 13, 10933- 10943.
2	Corbett A T, Anderson J R.. Knowledge tracing: modeling the acquisition of procedural knowledge. User Modeling and User-Adapted Interaction, 1994, 4 (4): 253- 278.
3	Cen H, Koedinger K, Junker B. Comparing two IRT models for conjunctive skills [C]//Woolf B P, Aïmeur E, Nkambou R, et al. Intelligent Tutoring Systems. Berlin, Heidelberg: Springer, 2008: 796-798.
4	Vie J J, Kashima H. Knowledge tracing machines: factorization machines for knowledge tracing [C]//Proceedings of the AAAI Conference on Artificial Intelligence. 2019, 33(1): 750-757.
5	Rendle S. Factorization machines [C]//2010 IEEE International conference on data mining. IEEE, 2010: 995-1000.
6	Piech C, Bassen J, Huang J, et al. Deep knowledge tracing [C]//Advances in Neural Information Processing Systems. New York: Curran Associates, Inc., 2015, 28: 505-513.
7	Zhang J N, Shi X J, King I, et al. Dynamic key-value memory networks for knowledge tracing [C]//Proceedings of the 26th International Conference on World Wide Web. International World Wide Web Conferences Steering Committee, 2017: 765-774.
8	Pandey S, Karypis G. A self-attentive model for knowledge tracing [PP/OL]. arXiv (2019-07-16)[2026-07-14]. https://doi.org/10.48550/arXiv.1907.06837.
9	Ghosh A, Heffernan N, Lan A S. Context-aware attentive knowledge tracing [C]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. ACM, 2020: 2330-2339.
10	王畅, 马丹, 许华容, 等.. SA-MGKT: 基于自注意力融合的多图知识追踪方法. 华东师范大学学报(自然科学版), 2024 (5): 20- 31.
11	Lee W, Chun J, Lee Y, et al. Contrastive learning for knowledge tracing [C]//Proceedings of the ACM Web Conference 2022. ACM, 2022: 2330-2338.
12	Cui J J, Yu M H, Jiang B, et al. Interpretable knowledge tracing via response influence-based counterfactual reasoning [C]//2024 IEEE 40th International Conference on Data Engineering (ICDE). IEEE, 2024: 1103-1116.
13	Scarselli F, Gori M, Tsoi A C, et al.. The graph neural network model. IEEE Transactions on Neural Networks, 2009, 20 (1): 61- 80.
14	Yang Y, Shen J, Qu Y R, et al. GIKT: a graph-based interaction model for knowledge tracing [C]//Hutter F, Kersting K, Lijffijt J, et al. Machine Learning and Knowledge Discovery in Databases. Cham: Springer International Publishing, 2021: 299-315.
15	Kipf T N, Welling M. Semi-supervised classification with graph convolutional networks [PP/OL]. V4. arXiv (2017-02-22) [2026-07-14]. https://doi.org/10.48550/arXiv.1609.02907.
16	Nakagawa H, Iwasawa Y, Matsuo Y. Graph-based knowledge tracing: modeling student proficiency using graph neural network [C]//2019 IEEE/WIC/ACM International Conference on Web Intelligence (WI). IEEE, 2019: 156-163.
17	Wang M D, Peng C, Yang R, et al. GASKT: a graph-based attentive knowledge-search model for knowledge tracing [C]//Qiu H, Zhang C, Fei Z M, et al. Knowledge Science, Engineering and Management. Cham: Springer International Publishing, 2021: 268-279.
18	Schlichtkrull M, Kipf T N, Bloem P, et al. Modeling relational data with graph convolutional networks [C]//Gangemi A, Navigli R, Vidal M E, et al. The Semantic Web. Cham: Springer International Publishing, 2018: 593-607.
19	Lipton Z C, Berkowitz J, Elkan C. A critical review of recurrent neural networks for sequence learning [PP/OL]. V4. arXiv (2015-10-17) [2026-07-14]. https://doi.org/10.48550/arXiv.1506.00019.
20	Choi Y, Lee Y, Cho J, et al. Towards an appropriate query, key, and value computation for knowledge tracing [C]//Proceedings of the Seventh ACM Conference on Learning @ Scale. ACM, 2020: 341-344.
21	Xu B H, Huang Z Y, Liu J Y, et al. Learning behavior-oriented knowledge tracing [C]//Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. ACM, 2023: 2789-2800.
22	Nagatani K, Zhang Q, Sato M, et al. Augmenting knowledge tracing by considering forgetting behavior [C]//The World Wide Web Conference. ACM, 2019: 3101-3107.
23	Liu Z T, Liu Q Q, Chen J H, et al. PyKT: a Python library to benchmark deep learning based knowledge tracing models [C]//Advances in Neural Information Processing Systems 35. Neural Information Processing Systems Foundation, Inc. (NeurIPS), 2022: 18542-18555.
24	Yeung C K, Yeung D Y. Addressing two problems in deep knowledge tracing via prediction-consistent regularization [C]//Proceedings of the Fifth Annual ACM Conference on Learning at Scale. ACM, 2018: 1-10.
25	Tong S W, Liu Q, Huang W, et al. Structure-based knowledge tracing: an influence propagation view [C]//2020 IEEE International Conference on Data Mining (ICDM). IEEE, 2020: 541-550.
26	Yeung C K. Deep-IRT: Make deep learning based knowledge tracing explainable using item response theory [PP/OL]. arXiv (2019-04-26) [2026-07-14]. https://doi.org/10.48550/arXiv.1904.11738.
27	Liu Z T, Liu Q Q, Chen J H, et al. SimpleKT: a simple but tough-to-beat baseline for knowledge tracing [C]//International Conference on Learning Representations. 2023.

数据集	学生数量/个	问题数量/个	知识概念数量/个	交互记录数量/条	稀疏度/%
Assist09	4163	17751	123	283105	99.62
Assist12	25461	42992	258	2681639	99.77
Junyi	175359	716	39	25797740	98.25

数据集	知识概念平均问题数量/个	问题平均知识概念数量/个
Assist09	172.732	1.197
Assist12	166.636	1.000
Junyi	18.359	1.000

数据集	答题正确率/%	答题错误率/%	平均尝试次数/次	平均提示次数/次
Assist09	65.81	34.19	1.600	0.451
Assist12	69.59	30.41	1.355	0.396
Junyi	82.92	17.08	1.363	0.281

模型	Assist09		Assist12		Junyi
模型	$ \mathcal{U} $	$ {\mathcal{U}}_{\text{D}} $	$ \mathcal{U} $	$ {\mathcal{U}}_{\text{D}} $	$ \mathcal{U} $	$ {\mathcal{U}}_{\text{D}} $
DKT	0.7715	0.6912	0.7328	0.6645	0.7462	0.6781
DKT+	0.7775	0.6968	0.7543	0.6751	0.7503	0.6817
GIKT	0.7863	0.717 1	0.7612	0.6908	0.7576	0.699 2
GKT	0.7815	0.7097	0.7589	0.6885	0.7496	0.6905
SKT	0.7787	0.7071	0.7572	0.6870	0.7431	0.6946
DKVMN	0.7712	0.7009	0.7598	0.6894	0.7559	0.6975
Deep-IRT	0.7758	0.7052	0.7547	0.6855	0.7502	0.6916
SAKT	0.7624	0.6923	0.7483	0.6796	0.7357	0.6872
SAINT	0.7631	0.6930	0.7491	0.6803	0.7364	0.6879
AKT	0.789 2	0.6971	0.770 5	0.6795	0.757 8	0.6894
SimpleKT	0.7827	0.7008	0.7662	0.6755	0.7528	0.6896
CL4KT	0.7721	0.7118	0.7621	0.691 6	0.7382	0.6941
HGC-FKT	0.8108	0.7796	0.7912	0.7581	0.7658	0.7345

模型	Assist09		Assist12		Junyi
模型	$ \mathcal{U} $	$ {\mathcal{U}}_{\text{D}} $	$ \mathcal{U} $	$ {\mathcal{U}}_{\text{D}} $	$ \mathcal{U} $	$ {\mathcal{U}}_{\text{D}} $
DKT	0.7293	0.6482	0.7185	0.6610	0.7246	0.6891
DKT+	0.7316	0.6505	0.7279	0.6698	0.7271	0.6916
GIKT	0.7329	0.686 1	0.7256	0.692 2	0.7243	0.7093
GKT	0.7306	0.6835	0.7231	0.6896	0.7229	0.7082
SKT	0.7275	0.6802	0.7248	0.6910	0.7237	0.7090
DKVMN	0.7389	0.6632	0.7342	0.6765	0.734 9	0.7014
Deep-IRT	0.7301	0.6553	0.7359	0.6780	0.7328	0.6993
SAKT	0.7273	0.6461	0.7235	0.6658	0.7233	0.6880
SAINT	0.7164	0.6372	0.7102	0.6534	0.7091	0.6743
AKT	0.741 8	0.6610	0.737 6	0.6792	0.7346	0.7010
SimpleKT	0.7325	0.6523	0.7284	0.6703	0.7279	0.6924
CL4KT	0.7264	0.6721	0.7227	0.6812	0.7224	0.709 8
HGC-FKT	0.7477	0.7123	0.7431	0.7156	0.7422	0.7245