arXiv:2604.21495cs.LGcs.AI2026-04中稿 · TACL

用结构提示和自监督学习提升表格数值推理的跨领域泛化能力

Generalizing Numerical Reasoning in Table Data through Operation Sketches and Self-Supervised Learning

  • 通过操作草图和标题匿名化,引导模型关注结构而非关键词
  • 仅用10%数据训练即达80.13%准确率,超越全量数据基线
  • 适合需要跨领域稳定推理的金融、医疗等专业表格场景

专家领域表格中的数值推理常表现出高域内准确率但跨领域鲁棒性差。基于监督微调(SFT)的模型易依赖表头-操作捷径,而非结构化推理。本文提出持续预训练框架TaNOS,包含三项设计:(i) 标题匿名化以减少词汇记忆,(ii) 操作草图提供最小结构提示,(iii) 程序优先的自监督预训练,从给定表格构建保证正确性的程序-问题对。通过解耦领域语义与数值操作结构,TaNOS提升了数值推理的迁移能力。应用于80亿参数指令微调模型,在仅使用10%训练数据的情况下,FinQA上执行准确率达80.13%,优于使用全量数据的SFT基线(73.97%)及GPT-5、Gemini-2.5-Pro等专有模型。在跨领域实验中,TaNOS的跨域差距不足2个百分点,而标准SFT超过10个百分点。结果表明,操作草图、无标题依赖表示与保证正确的自监督机制可显著增强数值推理的跨领域稳健性。

原文摘要 · Abstract (English)

Numerical reasoning over expert-domain tables often exhibits high in-domain accuracy but limited robustness to domain shift. Models trained with supervised fine-tuning (SFT) on specific datasets tend to rely on header-operation shortcuts rather than structural reasoning. We introduce TaNOS, a continual pre-training framework comprising three components: (i) header anonymization to reduce lexical memorization, (ii) operation sketches that provide minimal structural cues, and (iii) self-supervised pretraining that constructs correctness-guaranteed program-question pairs from given tables in a program-first manner. By decoupling domain semantics and numerical operation structure, TaNOS improves the transferability of numerical reasoning. Applied to an 8B instruction-tuned model, TaNOS achieves 80.13% execution accuracy on FinQA with only 10% train data, outperforming SFT baseline (73.97%) with full train data and proprietary models such as GPT-5, Gemini-2.5-Pro. Furthermore, in the domain-shift experiments, TaNOS displays nearly-negligible cross-domain gap (<2pp) when standard SFT shows over 10pp gap. These results suggest that structural guidance with operation sketches, header-agnostic representations, and correctness-guaranteed self-supervision can improve the robustness of numerical reasoning across diverse expert-domain tables.

数值推理表格理解自监督学习跨领域泛化

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。