迈向稳健的数值声明验证
Towards Robust Numerical Claim Verification
浏览论文内容
中文总结 AI 辅助
针对LLM数值声明验证的脆弱性，通过对抗性微调小型Qwen3模型，在标签翻转扰动上达到98.7%准确率，并泛化至未见扰动和跨语言场景。
中文摘要 AI 辅助
大型语言模型（LLMs）广泛用于声明验证，但在数值推理方面仍然脆弱：即使数值的微小变化也会急剧降低准确率。我们表明，这种脆弱性在前沿LLMs中依然存在，但可以通过对数值扰动示例进行对抗性微调来缓解。使用参数高效微调，小型Qwen3模型（0.6B–8B）在标签翻转扰动上达到98.7%的准确率，优于更大的零样本模型和前沿系统（GPT-5.4 Pro（74.0%）和Gemini 2.5 Flash（73.9%））。这些增益泛化到未见过的扰动类型，表明模型学到了稳健的数值决策边界而非记忆编辑。鲁棒性还能在没有目标域数据的情况下迁移，显著提高了西班牙语的跨语言性能。我们进一步表明，相同的微调方法也能赋予对证据侧扰动的鲁棒性，使用VitaminC数据集。
英文摘要
Large language models (LLMs) are widely used for claim verification, yet remain brittle for numerical reasoning: even small changes in value can sharply degrade accuracy. We show that this brittleness persists in frontier LLMs, but can be mitigated through adversarial fine-tuning on numerically perturbed examples. Using parameter-efficient fine-tuning, small Qwen3 models (0.6B$\unicode{x2013}$8B) reach 98.7% accuracy on label-flipping perturbations, outperforming larger zero-shot models and frontier systems (GPT-5.4 Pro (74.0%) and Gemini 2.5 Flash (73.9%)). The gains generalise to unseen perturbation types, indicating robust numerical decision boundaries rather than memorised edits. Robustness also transfers without target-domain data, significantly improving cross-lingual performance in Spanish. We further show that the same fine-tuning recipe confers robustness to evidence-side perturbations, using the VitaminC dataset.
发表机构
- University of Stavanger（斯塔万格大学）
- Factiverse AI（Factiverse AI公司）
机构由 AI 辅助整理，请以论文原文为准。
