DocTrace: Towards Traceable Long Document VQA via Hierarchical Evidence Graph Reasoning

📄 arXiv: 2608.03292v1 📥 PDF

作者: Le Xiang, Zhicheng Guan, Hong Chen, Xiaocong Lin, Zhenghua Lei, Teng Hu, Bolei He, Long Zeng

分类: cs.AI

发布日期: 2026-08-04

💡 一句话要点

提出DocTrace以解决长文档VQA的可追溯性问题

🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 长文档视觉问答 证据图推理 多模态大语言模型 透明推理 信息检索

📋 核心要点

  1. 现有的长文档视觉问答方法缺乏明确的证据表示和验证机制，导致答案准确性和可追溯性不足。
  2. 本文提出DocTrace，通过分层证据图推理来显式化证据来源，增强长文档理解的透明性。
  3. 实验结果表明，DocTrace在MMLongBench-Doc、LongDocURL和SlideVQA等基准上均显著超越现有方法，提升幅度达到14.4、11.3和11.7分。

📝 摘要（中文）

长文档视觉问答（LongDocVQA）要求多模态大语言模型（MLLMs）在多个页面上定位、整合和推理异构文档元素。现有方法缺乏明确的机制来表示和验证推理过程中如何逐步构建证据，限制了答案的准确性和可追溯性。本文将LongDocVQA视为显式证据图推理问题，提出了DocTrace，一个分层框架，逐步执行证据定位、结构化文档解析和证据图推理，以实现显式证据来源。通过两阶段训练框架，DocTrace在多个基准测试中表现优异，超越了现有的开源基线和专有MLLMs。

🔬 方法详解

问题定义：本文旨在解决长文档视觉问答中的证据可追溯性问题。现有方法在推理过程中缺乏对证据来源的明确表示，导致答案的准确性和透明性不足。

核心思路：DocTrace的核心思路是将长文档视觉问答视为显式证据图推理问题，通过分层框架逐步进行证据定位和推理，从而实现对证据来源的明确追踪。

技术框架：DocTrace的整体架构包括证据定位、结构化文档解析和证据图推理三个主要模块。首先进行证据的定位，然后解析文档结构，最后通过证据图进行推理。

关键创新：DocTrace的主要创新在于构建了可追溯的证据图，提供了节点级的来源信息，与现有方法的隐式推理方式形成鲜明对比。

关键设计：在训练过程中，采用了两阶段的训练框架，首先进行联合监督微调（SFT），然后通过任务特定的群体相对策略优化（GRPO）进一步优化能力，设计了专门的奖励机制以提升模型性能。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

DocTrace在多个基准测试中表现出色，相较于Qwen3-VL-8B-Instruct骨干网络，分别在MMLongBench-Doc、LongDocURL和SlideVQA上实现了14.4、11.3和11.7的绝对提升，展示了其在长文档视觉问答中的优越性能。

🎯 应用场景

DocTrace的研究成果在长文档理解、信息检索和智能问答系统等领域具有广泛的应用潜力。其可追溯的证据图结构能够为用户提供透明的推理过程，增强系统的可信度和实用性，未来可在教育、法律和医疗等行业中发挥重要作用。

📄 摘要（原文）

Long Document Visual Question Answering (LongDocVQA) requires Multimodal Large Language Models (MLLMs) to locate, integrate, and reason over heterogeneous document elements distributed across multiple pages. Existing approaches, including end-to-end MLLMs, retrieval-augmented generation (RAG) pipelines, and document agents, often lack explicit mechanisms to represent and verify how grounded evidence is progressively composed during reasoning, limiting both answer accuracy and traceability. In this paper, we cast LongDocVQA as an explicit evidence graph reasoning problem rather than implicit answer prediction. To this end, we propose DocTrace, a hierarchical framework that progressively performs evidence localization, structured document parsing, and evidence graph reasoning to enable explicit evidence provenance. To effectively learn these capabilities, we develop a two-stage training framework: joint Supervised Fine-Tuning (SFT) first initializes evidence localization and graph reasoning abilities, followed by task-specific Group Relative Policy Optimization (GRPO) with dedicated rewards to further optimize these capabilities. Extensive experiments on MMLongBench-Doc, LongDocURL, and SlideVQA demonstrate that DocTrace consistently outperforms both existing open-source baselines and proprietary MLLMs. Compared with the Qwen3-VL-8B-Instruct backbone, DocTrace achieves absolute improvements of 14.4, 11.3, and 11.7 points on the three benchmarks, respectively. Beyond competitive performance, DocTrace constructs traceable evidence graphs with explicit node-level provenance, enabling transparent and verifiable reasoning for long document understanding.