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 2511.03047cs.LG
收藏

多轮目标驱动交互的无监督评估

Unsupervised Evaluation of Multi-Turn Objective-Driven Interactions

Department of Electrical Engineering Stanford University(电气工程系 斯坦福大学);Emissary Technologies

Emi Soroka, Tanmay Chopra, Krish Desai, Sanjay Lall

AI总结:

针对多轮目标驱动交互难以评估的问题,提出首套无监督评估指标,利用无标签数据统计特性与微调LLMs适应分布偏移,实现用户目标标注、目标完成度衡量及不确定性量化,并在开放域和特定任务数据上验证有效性。

摘要:

大型语言模型(LLMs)在企业应用中日益普及,在这些应用中,AI agents 和人类进行目标驱动交互。然而,这些系统难以评估:数据可能复杂且无标签;大规模人工标注通常不切实际;自定义指标能监控特定错误,但无法监控以前未检测到的错误;且 LLM judges 可能产生不可靠的结果。我们引入了第一套用于目标驱动交互的无监督指标,利用无标签交互数据的统计特性,并使用微调的 LLMs 来适应分布偏移。我们开发了用于标注用户目标、衡量目标完成情况以及量化 LLM 不确定性的指标,而无需将评估建立在人类生成的理想响应之上。我们的方法在开放域和特定任务交互数据上得到了验证。

英文摘要:

Large language models (LLMs) have seen increasing popularity in enterprise applications where AI agents and humans engage in objective-driven interactions. However, these systems are difficult to evaluate: data may be complex and unlabeled; human annotation is often impractical at scale; custom metrics can monitor for specific errors, but not previously-undetected ones; and LLM judges can produce unreliable results. We introduce the first set of unsupervised metrics for objective-driven interactions, leveraging statistical properties of unlabeled interaction data and using fine-tuned LLMs to adapt to distributional shifts. We develop metrics for labeling user goals, measuring goal completion, and quantifying LLM uncertainty without grounding evaluations in human-generated ideal responses. Our approach is validated on open-domain and task-specific interaction data.

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。

↑