MathDoc: Benchmarking Structured Extraction and Active Refusal on Noisy Mathematics Exam Papers

TL;DR

MathDoc introduces the first document-level benchmark for noisy high school mathematics exams, targeting both structured extraction and active refusal when inputs are incomplete or illegible. The dataset comprises $3{,}609$ questions with real-world artifacts and an explicit unrecognizable subset, accompanied by a multi-dimensional evaluation framework for stem accuracy, visual similarity, and refusal capability. Across open- and closed-source SOTA multimodal LLMs, results reveal strong extraction performance on recognizable samples but a pervasive tendency to overfill or guess for incomplete inputs, highlighting a critical reliability gap. The work provides a rigorous benchmark, evaluation protocol, and initial strategies to improve refusal, aiming to advance robust document processing systems under degraded conditions.

Abstract

The automated extraction of structured questions from paper-based mathematics exams is fundamental to intelligent education, yet remains challenging in real-world settings due to severe visual noise. Existing benchmarks mainly focus on clean documents or generic layout analysis, overlooking both the structural integrity of mathematical problems and the ability of models to actively reject incomplete inputs. We introduce MathDoc, the first benchmark for document-level information extraction from authentic high school mathematics exam papers. MathDoc contains \textbf{3,609} carefully curated questions with real-world artifacts and explicitly includes unrecognizable samples to evaluate active refusal behavior. We propose a multi-dimensional evaluation framework covering stem accuracy, visual similarity, and refusal capability. Experiments on SOTA MLLMs, including Qwen3-VL and Gemini-2.5-Pro, show that although end-to-end models achieve strong extraction performance, they consistently fail to refuse illegible inputs, instead producing confident but invalid outputs. These results highlight a critical gap in current MLLMs and establish MathDoc as a benchmark for assessing model reliability under degraded document conditions. Our project repository is available at \href{https://github.com/winnk123/papers/tree/master}{GitHub repository}

Figures (15)

• Figure 1: Performance comparison of six Multimodal Large Language Models (MLLMs) across five metrics (Stem Accuracy, Visual Similarity, Refusal F1/Precision/Recall) in choice questions.
• Figure 2: (a) The process of Data Construction. (b) The distribution of all types of questions. (c) The average length of questions and their average formula ratio.
• Figure 3: The pipeline of evaluation process
• Figure 4: Model performance on Solve, sorted in descending order according to the Final metric (left to right).
• Figure 5: Model performance on Choice, sorted in descending order according to the Final metric (left to right).