让大模型生成答案分解步骤，提升长文档问答的可信溯源能力。
Decomposition-Enhanced Training for Post-Hoc Attributions In Language Models
- 将答案拆解为与原文对应的单元，作为推理中间步骤训练模型
- 在多跳、抽象、半抽取场景下，溯源准确率显著优于现有方法
- 适合需要高可信度回答溯源的研究者和应用开发者
大语言模型在长文档问答中的应用日益广泛，可靠的信息溯源对建立信任至关重要。现有后处理溯源方法在提取式问答中表现良好，但在多跳、抽象及半抽取场景中表现不佳，因答案需融合多个段落信息。本文提出将后处理溯源重构为推理问题，通过将答案分解为与特定上下文相关的组成部分来解决。实验表明，引导模型生成此类分解可提升性能。基于此，我们提出 DecompTune：一种后训练方法，使模型在生成答案时输出分解作为中间推理步骤。我们构建了一个包含复杂问答任务的多样化数据集，由强模型标注分解结果，并使用两阶段 SFT + GRPO 管道对 Qwen-2.5（7B 与 14B）进行微调，采用任务定制奖励。大量实验与消融分析显示，DecompTune 显著提升溯源质量，优于以往方法，达到或超越前沿模型水平。
原文摘要 · Abstract (English)
Large language models (LLMs) are increasingly used for long-document question answering, where reliable attribution to sources is critical for trust. Existing post-hoc attribution methods work well for extractive QA but struggle in multi-hop, abstractive, and semi-extractive settings, where answers synthesize information across passages. To address these challenges, we argue that post-hoc attribution can be reframed as a reasoning problem, where answers are decomposed into constituent units, each tied to specific context. We first show that prompting models to generate such decompositions alongside attributions improves performance. Building on this, we introduce DecompTune, a post-training method that teaches models to produce answer decompositions as intermediate reasoning steps. We curate a diverse dataset of complex QA tasks, annotated with decompositions by a strong LLM, and post-train Qwen-2.5 (7B and 14B) using a two-stage SFT + GRPO pipeline with task-specific curated rewards. Across extensive experiments and ablations, DecompTune substantially improves attribution quality, outperforming prior methods and matching or exceeding state-of-the-art frontier models.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。