arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2603.28005cs.CL
收藏

重新思考用于LLM裁判的原子分解:一种受参考影响的问答评估的提示控制研究

Atomic and Holistic LLM Judges for Reference-Grounded Support Labels: A Prompt-Controlled Comparison

University of California, Berkeley(加州大学伯克利分校)

Xinran Zhang

AI总结:

本文研究了参考导向问答评估中原子分解与整体裁判的性能差异,发现整体裁判在多数基准中表现更优,尤其在部分支持案例中表现突出,且参考质量下降对两种裁判方式均产生显著影响。

摘要:

原子分解——在验证每个候选答案前将其分解成声明——是LLM参考导向裁判中广泛采用的设计。然而,原子提示通常更丰富且更长,不清楚分解本身是否带来优势还是更丰富的提示。我们研究了这种设计在基准式完整性敏感的参考支持分类中的表现:根据提供的参考将候选答案分类为完全支持、部分支持或不支持。我们比较了自我分解的原子裁判(单提示分解和验证)与具有相同输入和类似详细评分标准的提示控制整体裁判。在TruthfulQA、ASQA和QAMPARI数据集上,每个数据集200个源示例,四个模型家族,源级配对测试,聚类Bootstrap,以及每个设计家族三个冻结提示变体的聚合,我们发现整体裁判在三个基准中的两个上表现等于或优于原子裁判:ASQA和QAMPARI在所有四个家族中偏整体(统计上可靠在三个中),而TruthfulQA显示了微小的原子优势。整体优势集中在部分支持案例——不完整性检测。对人类注释的敏感性检查确认了在基准完整性和人类事实正确性标准下的排名。我们的发现特定于三个QA风格基准,每个基准200个源示例,自我分解单提示模式;多阶段原子管道和非QA任务仍未经测试。在检查的扰动中,参考质量下降对两个裁判家族产生了最大的准确性下降。

英文摘要:

When an LLM judge only has to assign a three-way support label to a candidate answer given a reference, does asking it to decompose the answer into atomic claims help, and at what cost? We compare four single-call designs that share the judge model, the inputs, and the level of instruction detail: candidate-side atomic decomposition, a matched holistic rubric, reference-side decomposition that checks whether each reference claim is covered, and a bidirectional combination. Support labels are constructed from TruthfulQA, ASQA, and QAMPARI references (200 questions and 400 rows per dataset). All four designs are run with Opus-4.6, GPT-4.1, and Gemini Flash Lite; Sonnet-4.6 is added for the candidate-side and holistic designs. Candidate-side decomposition is weak where the label depends on completeness: the holistic rubric is more accurate on ASQA and QAMPARI for every judge while using fewer tokens. Reference-side decomposition is 12.5-21.3 points more accurate than holistic on ASQA for 28-29% more tokens, and stays near the holistic ceiling on QAMPARI at 63-69% more. On TruthfulQA misconceptions, candidate-side decomposition is competitive and significantly better for two judges. On a 60-row single-author subset whose labels are looser than strict reference completeness, candidate-side decomposition edges ahead of holistic for every judge, reversing the construction-label order. What a judge decomposes should follow what the label measures.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑