Zhangqi Duan

Zhangqi Duan

PhD Student at UMass Amherst

Publications

You can also find my articles on my Google Scholar profile.

Conference Papers

KASER: Knowledge-Aligned Student Error Simulator for Open-Ended Coding Tasks

Published in ACL, 2026

When applying LLMs for student code simulation, we find that both prompt-based and SFT approaches tend to generate overly correct solutions, rather than reflecting the student’s actual knowledge and likely mistakes. To enable more realistic student simulation, we introduce KASER, which is trained with GRPO and a carefully designed reward that balances both diversity and error coverage.

Recommended citation: @misc{duan2026kaserknowledgealignedstudenterror, title={KASER: Knowledge-Aligned Student Error Simulator for Open-Ended Coding Tasks}, author={Zhangqi Duan and Nigel Fernandez and Andrew Lan}, year={2026}, eprint={2601.06633}, archivePrefix={arXiv}, primaryClass={cs.LG}, url={https://arxiv.org/abs/2601.06633}, }
Download Paper

Automated Knowledge Component Generation for Interpretable Knowledge Tracing in Coding Problems

Published in ACL Findings, 2026

We introduce KCGen-KT, a controllable framework for generating Knowledge Components (KCs) at varying granularity levels based on instructional needs. We also propose a soft token conversion mechanism that allows gradient flow across models (knowledge estimator and code generator), enabling more interpretable training.

Recommended citation: @misc{duan2025automatedknowledgecomponentgeneration, title={Automated Knowledge Component Generation for Interpretable Knowledge Tracing in Coding Problems}, author={Zhangqi Duan and Nigel Fernandez and Arun Balajiee Lekshmi Narayanan and Mohammad Hassany and Rafaella Sampaio de Alencar and Peter Brusilovsky and Bita Akram and Andrew Lan}, year={2025}, eprint={2502.18632}, archivePrefix={arXiv}, primaryClass={cs.AI}, url={https://arxiv.org/abs/2502.18632}, }
Download Paper

Using LLMs for Knowledge Component-level Correctness Labeling in Open-ended Coding Problems

Published in AIED, 2026

This paper is the first work to leverage LLM for Knowledge Components (KC) level correctness labeling on coding problem.

Recommended citation: @misc{duan2026usingllmsknowledgecomponentlevel, title={Using LLMs for Knowledge Component-level Correctness Labeling in Open-ended Coding Problems}, author={Zhangqi Duan and Arnav Kankaria and Dhruv Kartik and Andrew Lan}, year={2026}, eprint={2602.17542}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2602.17542}, }
Download Paper

Test Case-Informed Knowledge Tracing for Open-ended Coding Tasks

Published in LAK, 2025

This paper is about a new defined KT task: Test Case Outcome Prediction

Recommended citation: @misc{duan2024testcaseinformedknowledgetracing, title={Test Case-Informed Knowledge Tracing for Open-ended Coding Tasks}, author={Zhangqi Duan and Nigel Fernandez and Alexander Hicks and Andrew Lan}, year={2024}, eprint={2410.10829}, archivePrefix={arXiv}, primaryClass={cs.CY}, url={https://arxiv.org/abs/2410.10829}, }
Download Paper

Multilingualism, Transnationality, and K-pop in the Online# StopAsianHate Movement

Published in WebSci, 2025

Recommended citation: @misc{masis2025multilingualismtransnationalitykpoponline, title={Multilingualism, Transnationality, and K-pop in the Online #StopAsianHate Movement}, author={Tessa Masis and Zhangqi Duan and Weiai Wayne Xu and Ethan Zuckerman and Jane Yeahin Pyo and Brendan O'Connor}, year={2025}, eprint={2503.02707}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2503.02707}, }
Download Paper

README: Bridging Medical Jargon and Lay Understanding for Patient Education through Data-Centric NLP

Published in EMNLP Findings, 2024

Recommended citation: @misc{yao2024readmebridgingmedicaljargon, title={README: Bridging Medical Jargon and Lay Understanding for Patient Education through Data-Centric NLP}, author={Zonghai Yao and Nandyala Siddharth Kantu and Guanghao Wei and Hieu Tran and Zhangqi Duan and Sunjae Kwon and Zhichao Yang and README annotation team and Hong Yu}, year={2024}, eprint={2312.15561}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2312.15561}, }
Download Paper