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 2503.05965cs.LGcs.CYcs.HC
收藏

验证评分不确定条件下的LLM-as-a-Judge系统

Validating LLM-as-a-Judge Systems under Rating Indeterminacy

Carnegie Mellon University(卡内基梅隆大学);Microsoft Research(微软研究院)

Luke Guerdan, Solon Barocas, Kenneth Holstein, Hanna Wallach, Zhiwei Steven Wu, Alexandra Chouldechova

AI总结:

针对评分不确定性问题,提出多标签响应集评分框架,避免强制选择导致的验证偏差,实验表明该方法选出的法官系统性能提升达31%。

摘要:

LLM-as-a-judge范式,即由法官LLM系统替代人类评分员对其他生成式AI(GenAI)系统的输出进行评分,在扩展和标准化GenAI评估中发挥着关键作用。为了验证此类法官系统,评估者首先为验证语料库中的每个项目收集多个人类评分,然后将评分聚合为每个项目的单一黄金标签评分,以此评估人类与法官之间的一致性。然而,对于许多项目,评分标准可能允许多种有效解释,因此人类或LLM评分员可能认为多个评分“合理”或“正确”。我们将这种情况称为评分不确定性。问题在于,许多包含评分不确定性的评分任务依赖于强制选择引出法,即要求评分员为每个项目仅选择一个评分。在本文中,我们提出了一个在评分不确定性条件下验证LLM-as-a-judge系统的框架。我们在不同人类-法官一致性度量下的法官系统性能度量之间,以及不同评分引出和聚合方案之间,建立了理论联系。我们证明,在响应强制选择评分指令时,人类和LLM解决评分不确定性的方式差异可能严重偏倚LLM-as-a-judge的验证结果。通过涉及11个真实世界评分任务和9个商业LLM的大量实验,我们表明,依赖强制选择评分的标准验证方法所选择的法官系统高度次优,其性能比我们采用多标签“响应集”评分以考虑评分不确定性的方法所选择的法官系统差多达31%。最后,我们为更规范的LLM-as-a-judge验证方法提出了具体建议。

英文摘要:

The LLM-as-a-judge paradigm, in which a judge LLM system replaces human raters in rating the outputs of other generative AI (GenAI) systems, plays a critical role in scaling and standardizing GenAI evaluations. To validate such judge systems, evaluators assess human--judge agreement by first collecting multiple human ratings for each item in a validation corpus, then aggregating the ratings into a single, per-item gold label rating. For many items, however, rating criteria may admit multiple valid interpretations, so a human or LLM rater may deem multiple ratings "reasonable" or "correct." We call this condition rating indeterminacy. Problematically, many rating tasks that contain rating indeterminacy rely on forced-choice elicitation, whereby raters are instructed to select only one rating for each item. In this paper, we introduce a framework for validating LLM-as-a-judge systems under rating indeterminacy. We draw theoretical connections between different measures of judge system performance under different human--judge agreement metrics, and different rating elicitation and aggregation schemes. We demonstrate that differences in how humans and LLMs resolve rating indeterminacy when responding to forced-choice rating instructions can heavily bias LLM-as-a-judge validation. Through extensive experiments involving 11 real-world rating tasks and 9 commercial LLMs, we show that standard validation approaches that rely upon forced-choice ratings select judge systems that are highly suboptimal, performing as much as 31% worse than judge systems selected by our approach that uses multi-label "response set" ratings to account for rating indeterminacy. We conclude with concrete recommendations for more principled approaches to LLM-as-a-judge validation.

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。

↑