Selected work

Representative Papers

Tracing Mathematical Proficiency Through Problem-Solving Processes

Nov 28, 2025

Traditional knowledge tracing (KT) methods rely solely on binary response correctness, failing to capture the latent structural composition of students’ mathematical competencies and suffering from severe interpretability limitations. Method: We propose KT-PSP, the first problem-solving process-driven KT framework. It employs a teacher–student–teacher large language model pipeline to automatically extract multidimensional mathematical proficiency signals from students’ step-by-step solution traces. We accompany this with KT-PSP-25—a newly released dataset covering 25 fine-grained mathematical competency dimensions and annotated solution processes. Our approach integrates procedural sequence modeling, task-adaptive proficiency metric design, and an intrinsic interpretability evaluation mechanism. Contribution/Results: Experiments demonstrate that KT-PSP significantly outperforms state-of-the-art baselines in prediction accuracy. Crucially, it enables dimension-aware knowledge state attribution and diagnostic feedback, establishing a novel, empirically grounded paradigm for interpretable KT.

0 citationsRead paper

Explain with Visual Keypoints Like a Real Mentor! A Benchmark for Multimodal Solution Explanation

Apr 04, 2025

Current large language models (LLMs) lack support for visual explanations in mathematical reasoning, despite the critical role of diagrams, auxiliary lines, and other visual aids in human pedagogy. To address this gap, we introduce *Visual Solution Explanation*—a novel multimodal task requiring joint generation of text explanations and corresponding visual elements (e.g., auxiliary lines, annotations, geometric constructions) that are semantically aligned and mutually reinforcing. We present MathExplain, the first education-oriented multimodal benchmark for this task, comprising 997 high-quality math problems, each annotated with fine-grained visual keypoints and aligned natural-language explanations. Extensive experiments reveal systematic deficiencies in open-source LLMs for vision–language collaborative reasoning, while proprietary multimodal models demonstrate nascent capability. All code and data are publicly released to advance explainable, pedagogically grounded AI for education.

0 citationsRead paper

Recent publications

Latest Papers

Tracing Mathematical Proficiency Through Problem-Solving Processes

Nov 28, 2025

Traditional knowledge tracing (KT) methods rely solely on binary response correctness, failing to capture the latent structural composition of students’ mathematical competencies and suffering from severe interpretability limitations. Method: We propose KT-PSP, the first problem-solving process-driven KT framework. It employs a teacher–student–teacher large language model pipeline to automatically extract multidimensional mathematical proficiency signals from students’ step-by-step solution traces. We accompany this with KT-PSP-25—a newly released dataset covering 25 fine-grained mathematical competency dimensions and annotated solution processes. Our approach integrates procedural sequence modeling, task-adaptive proficiency metric design, and an intrinsic interpretability evaluation mechanism. Contribution/Results: Experiments demonstrate that KT-PSP significantly outperforms state-of-the-art baselines in prediction accuracy. Crucially, it enables dimension-aware knowledge state attribution and diagnostic feedback, establishing a novel, empirically grounded paradigm for interpretable KT.

0 citationsRead paper

Explain with Visual Keypoints Like a Real Mentor! A Benchmark for Multimodal Solution Explanation

Apr 04, 2025

Current large language models (LLMs) lack support for visual explanations in mathematical reasoning, despite the critical role of diagrams, auxiliary lines, and other visual aids in human pedagogy. To address this gap, we introduce *Visual Solution Explanation*—a novel multimodal task requiring joint generation of text explanations and corresponding visual elements (e.g., auxiliary lines, annotations, geometric constructions) that are semantically aligned and mutually reinforcing. We present MathExplain, the first education-oriented multimodal benchmark for this task, comprising 997 high-quality math problems, each annotated with fine-grained visual keypoints and aligned natural-language explanations. Extensive experiments reveal systematic deficiencies in open-source LLMs for vision–language collaborative reasoning, while proprietary multimodal models demonstrate nascent capability. All code and data are publicly released to advance explainable, pedagogically grounded AI for education.

0 citationsRead paper