arXiv cs.CL· Yong Zu, Ninghua Yang, Zezhong Tan, Xinhong Ma, Ziqiang Dong, Yang Yang·· 2 天前AI 评分20
P2TPO：面向全页文档解析的结构对齐页面到 token 信用分配
P2TPO: Structure-Aligned Page-to-Token Credit Assignment for Full-Page Document Parsing
AI 导读
针对全页文档解析中页面级优势无法定位局部错误的问题，研究者提出结构对齐的信用分配框架 P2TPO，通过参考元素对齐、Unified Anchor Pricing（UAP）和 Span Credit Redistribution（SCR）将块级信用细化到 token 级。在六个文档解析基准上，P2TPO 的种子平均分超过所有对比的策略优化基线，并在不同模型规模和初始化下均取得增益。
正文 · 原文
Abstract:End-to-end full-page document parsing must preserve the content and structure of heterogeneous elements, yet otherwise accurate outputs can contain local errors and omissions. A single page-level advantage does not explicitly identify which elements differ in quality across sampled outputs or where their errors occur. We introduce Page-to-Token Policy Optimization (P2TPO), a structure-aligned credit-assignment framework that operates on full-page policy trajectories. Reference-element alignment first associates block-level quality contrasts with generated spans. Unified Anchor Pricing (UAP) incorporates matched and missing element outcomes and provides feedback relative to a fixed quality ceiling when sampled outputs receive identical but imperfect scores for a reference element. Within matched blocks, Span Credit Redistribution (SCR) redistributes existing block credit according to aligned correct and erroneous spans, using boundary proxies for within-block omissions while preserving the total block-derived advantage across its tokens. Under a fixed-weight additive page utility, we show that element-indexed contrasts retain local quality differences that page-level aggregation can cancel out. In the main experimental comparison, P2TPO outperforms all evaluated policy optimization baselines in seed-averaged scores on each of the six document parsing benchmarks. Additional experiments demonstrate gains across model scales and initializations.
| Comments: | 34 pages, 9 figures, including appendices |
| Subjects: | Machine Learning (cs.LG); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV) |
| Cite as: | arXiv:2610.21979 [cs.LG] |
| (or arXiv:2610.21979v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.21979 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yong Zu [view email]
[v1]
Thu, 15 Oct 2026 14:32:00 UTC (9,235 KB)
来源：arXiv cs.CL · arxiv.org