arXiv cs.AI· Lewei Xu, Yihao Ding, Zihan Xu, Daniel Yitian Su, Daochang Liu, Siwen Luo, Yifan Peng, Wei Liu·· 12 小时前AI 评分34
多页富视觉文档理解系统失败定位：一项实证归因研究
Locating Failure in Multi-Page Visually Rich Document Understanding Systems: An Empirical Attribution
AI 导读
研究提出实证归因框架，通过在统一文档问答流程中做受控干预，定位多页富视觉文档理解（MP-VRDU）系统的失败环节。在 MMLongBench-Doc 上的实验显示，图像可缓解解析器错误，提取文本则降低对视觉分辨率的依赖；加入相关非金标页面既能修复错误答案也会破坏正确答案，即使标注证据齐全，揭示出聚合准确率之下的不稳定性。弃答以减少无支撑回答换取误拒，其与推理指令的交互因骨干模型而异。
正文 · 原文
Abstract:Multi-page visually rich document understanding (MP-VRDU) systems can fail when evidence is not faithfully represented, not selected, or not used correctly, yet final accuracy alone does not reveal where failures occur. We present an empirical attribution framework that examines these failures through controlled interventions in a unified document question answering pipeline. Experiments on MMLongBench-Doc, with cross-dataset and cross-family checks, show that complementary modalities can reduce sensitivity to conversion errors: images mitigate parser errors, while extracted text reduces dependence on visual resolution. Adding related non-gold pages can both repair incorrect answers and break correct ones, even when all annotated evidence is present, revealing instability beneath aggregate accuracy. Abstention trades unsupported answers for false refusals, while its interaction with reasoning instructions differs across backbones. Qualitative examples and accuracy--cost comparisons connect these findings to guidance on representation, evidence selection, and reasoner choice under deployment constraints.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2608.07943 [cs.AI] |
| (or arXiv:2608.07943v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2608.07943 arXiv-issued DOI via DataCite |
Submission history
From: Lewei Xu [view email]
[v1]
Sat, 8 Aug 2026 06:07:55 UTC (1,121 KB)
[v2]
Thu, 15 Oct 2026 05:26:28 UTC (901 KB)
来源：arXiv cs.AI · arxiv.org