Elamur Research⌕
☰
home / papers / 2026 / Behavior-Aware Item Modeling via Dynamic Procedural Solution Representations for Knowledge Tracing

Behavior-Aware Item Modeling via Dynamic Procedural Solution Representations for Knowledge Tracing

Paper · 2026 · Jun Seo, Sangwon Ryu, Heejin Do et al.

Summary

Knowledge Tracing (KT) predicts whether a learner will answer a new item correctly based on past responses. Most KT work improves sequence modeling, while items are usually encoded as static vectors built from item–Knowledge Component (KC) structures. Such vectors capture declarative knowledge but not how a problem is actually solved. The authors propose Behavior-Aware Item Modeling (BAIM) to fill this gap. A reasoning language model (RLM) solves each item and splits its solution into Polya's four stages: understand, plan, carry out, and look back. Hidden states pooled across all RLM layers then yield one vector per stage. A context-conditioned router, driven by a GRU summary of the learner's history, selects the most relevant stage for each learner. The router is trained end-to-end inside the KT backbone, so the item encoder needs no separate pre-training. On the XES3G5M and NIPS34 math benchmarks, BAIM beats pretraining-based item embedding methods across five KT backbones. Gains are largest when learners attempt the same items repeatedly. This matters because item representations can adapt to each learner and remain useful when training data are scarce.

Key points

  • Backbones: BAIM was plugged into five KT backbones (AKT, qDKT, QIKT, simpleKT, sparseKT). Their sequence modeling and prediction layers were left unchanged.
  • Layer pooling: Pooling hidden states over the whole RLM embedding trajectory (layers 0–64) outperformed both final-layer-only representations and BERT-based text encoding.
  • Routing: Routing-based stage aggregation consistently beat fixed alternatives, such as holistic pooling, on XES3G5M.
  • Sample efficiency: With AKT, simpleKT, and sparseKT, BAIM trained on just 25% of XES3G5M students surpassed the Default item representation trained on the full dataset.
  • Limitations and availability: Evaluation covers only math datasets with public item content, so ASSISTments was excluded. Code and data are released at sjin4861/BAIM.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓ETH ZurichZurich · Switzerland72 works · 5,176 citations🎓Pohang University of Science and TechnologyPohang · South Korea7 works · 464 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksInformation Systems · 1,437 worksComputer Science Applications · 102 works

Cite

Cite as Jun Seo, Sangwon Ryu, Heejin Do, Hyounghun Kim and Gary Lee. 2026. Behavior-Aware Item Modeling via Dynamic Procedural Solution Representations for Knowledge Tracing. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1125

@inproceedings{seo2026behavior,
  title = {{Behavior-Aware Item Modeling via Dynamic Procedural Solution Representations for Knowledge Tracing}},
  author = {Jun Seo, Sangwon Ryu, Heejin Do, Hyounghun Kim and Gary Lee},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1125},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1125}
}

References

32 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. John R Anderson. 1996. Act: A simple theory of complex cognition. American psychologist, 51(4):355. find
  2. Shuai Bai, Yuxuan Cai, Ruizhe Chen, Keqin Chen, Xionghui Chen, Zesen Cheng, Lianghao Deng, Wei Ding, Chang Gao, Chunjiang Ge, Wenbin Ge, Zhifang Guo, Qidong Huang, Jie Huang, Fei Huang, Binyuan Hui, Shutong Jiang, Zhaohai Li, Mingsheng Li, and 45 others. 2025. Qwen3-vl technical report. Preprint, ar…
  3. Christopher M Bishop and Nasser M Nasrabadi. 2006. Pattern recognition and machine learning, volume 4. Springer. find
  4. Jiahao Chen, Zitao Liu, Shuyan Huang, Qiongqiong Liu, and Weiqi Luo. 2023. Improving interpretability of deep sequential knowledge tracing models with question-centric cognitive representations. In Proceedings of the AAAI conference on artificial intelligence, volume 37, pages 14196–14204. find
  5. Junyoung Chung, Caglar Gulcehre, KyungHyun Cho, and Yoshua Bengio. 2014. Empirical evaluation of gated recurrent neural networks on sequence modeling. arxiv 2014. arXiv preprint arXiv:1412.3555, 1412. find
  6. Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blistein, Ori Ram, Dan Zhang, Evan Rosen, and 1 others. 2025. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities. arX…
  7. Albert T Corbett and John R Anderson. 1994. Knowledge tracing: Modeling the acquisition of procedural knowledge. User modeling and user-adapted interaction, 4(4):253–278. find
  8. William Fedus, Barret Zoph, and Noam Shazeer. 2022. Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity. Journal of Machine Learning Research, 23(120):1–39. find
  9. Aritra Ghosh, Neil Heffernan, and Andrew S Lan. 2020. Context-aware attentive knowledge tracing. In Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining, pages 2330–2339. find
  10. Shuyan Huang, Zitao Liu, Xiangyu Zhao, Weiqi Luo, and Jian Weng. 2023. Towards robust knowledge tracing models via k-sparse attention. In Proceedings of the 46th international ACM SIGIR conference on research and development in information retrieval, pages 2441–2445. find
  11. Tao Huang, Xinjia Ou, Huali Yang, Shengze Hu, Jing Geng, Junjie Hu, and Zhuoran Xu. 2024. Remembering is not applying: Interpretable knowledge tracing for problem-solving processes. In Proceedings of the 32nd ACM International Conference on Multimedia, pages 3151–3159. find
  12. Ekaterina Krivich, Danial Hooshyar, Gustav Šír, Yeongwook Yang, Merja Bauters, Raija Hämäläinen, and Tommi Kärkkäinen. 2025. A systematic review of deep knowledge tracing (2015-2025): Toward responsible ai for education. find
  13. Unggi Lee, Sungjun Yoon, Joon Seo Yun, Kyoungsoo Park, YoungHoon Jung, Damji Stratton, and Hyeoncheol Kim. 2024. Difficulty-focused contrastive learning for knowledge tracing with a large language model-based difficulty prediction. In Proceedings of the 2024 Joint International Conference on Computa… find
  14. Yunfei Liu, Yang Yang, Xianyu Chen, Jian Shen, Haifeng Zhang, and Yong Yu. 2020. Improving knowledge tracing via pre-training question embeddings. In Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, IJCAI-20, pages 1577–1583. International Joint Conferences… find
  15. Zitao Liu, Qiongqiong Liu, Jiahao Chen, Shuyan Huang, and Weiqi Luo. 2023a. simpleKT: A simple but tough-to-beat baseline for knowledge tracing. In The Eleventh International Conference on Learning Representations. find
  16. Zitao Liu, Qiongqiong Liu, Teng Guo, Jiahao Chen, Shuyan Huang, Xiangyu Zhao, Jiliang Tang, Weiqi Luo, and Jian Weng. 2023b. Xes3g5m: A knowledge tracing benchmark dataset with auxiliary information. Advances in Neural Information Processing Systems, 36:32958–32970. find
  17. Yilmazcan Ozyurt, Stefan Feuerriegel, and Mrinmaya Sachan. 2024. Automated knowledge concept annotation and question representation learning for knowledge tracing. arXiv preprint arXiv:2410.01727. — [arXiv:2410.01727](https://arxiv.org/abs/2410.01727) find
  18. Chris Piech, Jonathan Bassen, Jonathan Huang, Surya Ganguli, Mehran Sahami, Leonidas J Guibas, and Jascha Sohl-Dickstein. 2015. Deep knowledge tracing. Advances in neural information processing systems, 28. find
  19. George Pólya. 1957. How to solve it: A new aspect of mathematical method, 2nd edition. Princeton university press, Princeton, NJ. find
  20. Alan H Schoenfeld. 2014. Mathematical problem solving. Elsevier. find
  21. Alan H Schoenfeld and Douglas J Herrmann. 1982. Problem perception and knowledge structure in expert and novice mathematical problem solvers. Journal of Experimental Psychology: Learning, Memory, and Cognition, 8(5):484. find
  22. Noam Shazeer, *Azalia Mirhoseini, *Krzysztof Maziarz, Andy Davis, Quoc Le, Geoffrey Hinton, and Jeff Dean. 2017. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer. In International Conference on Learning Representations. find
  23. Aaditya Singh, Adam Fry, Adam Perelman, Adam Tart, Adi Ganesh, Ahmed El-Kishky, Aidan McLaughlin, Aiden Low, AJ Ostrow, Akhila Ananthram, and 1 others. 2025. Openai gpt-5 system card. arXiv preprint arXiv:2601.03267.
  24. Oscar Skean, Md Rifat Arefin, Dan Zhao, Niket Nikul Patel, Jalal Naghiyev, Yann LeCun, and Ravid Shwartz-Ziv. 2025. Layer by layer: Uncovering hidden representations in language models. In Forty-second International Conference on Machine Learning. find
  25. Xiangyu Song, Jianxin Li, Qi Lei, Wei Zhao, Yunliang Chen, and Ajmal Mian. 2022. Bi-clkt: Bigraph contrastive learning based knowledge tracing. Knowledge-Based Systems, 241:108274. find
  26. Shashank Sonkar, Andrew E Waters, Andrew S Lan, Phillip J Grimaldi, and Richard G Baraniuk. 2020. qdkt: Question-centric deep knowledge tracing. arXiv preprint arXiv:2005.12442. — [arXiv:2005.12442](https://arxiv.org/abs/2005.12442) find
  27. John Sweller. 1988. Cognitive load during problem solving: Effects on learning. Cognitive science, 12(2):257–285. find
  28. Yixuan Tang and Yi Yang. 2024. Pooling and attention: What are effective designs for llm-based embedding models? arXiv preprint arXiv:2409.02727. — [arXiv:2409.02727](https://arxiv.org/abs/2409.02727) find
  29. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Ł ukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in Neural Information Processing Systems, volume 30. Curran Associates, Inc.
  30. Weiyun Wang, Zhangwei Gao, Lixin Gu, Hengjun Pu, Long Cui, Xingguang Wei, Zhaoyang Liu, Linglin Jing, Shenglong Ye, Jie Shao, and 1 others. 2025a. Internvl3.5: Advancing open-source multimodal models in versatility, reasoning, and efficiency. arXiv preprint arXiv:2508.18265.
  31. Wentao Wang, Huifang Ma, Yan Zhao, and Zhixin Li. 2024a. Pre-training question embeddings for improving knowledge tracing with self-supervised bi-graph co-contrastive learning. ACM Trans. Knowl. Discov. Data. find
  32. Notation Shape D input D kt D history f proj (·) W in W gate b gate r t−1 m t MLP p (·) 768 256 64 Linear(4D input → D kt ) + ReLU + Dropout ∈ R D history ×(2D kt ) ∈ R 4×(D kt +D history ) ∈ R 4 ∈ R D kt , r t−1 ∈ {0, 1} ∈ R D history Linear(D input → D input ) + ReLU + Dropout + Linear(D input → D… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/seo-2026-behavior-aware-item-modeling-dynamic/