On the Role of Citations in Preference Data
On the Role of Citations in Preference Data
关于引用在偏好数据中的作用
Abstract: Many NLP tasks require systems to provide attribution in their outputs—i.e. citations to grounding sources. Attribution serves as a bulwark against model hallucination and as a means for users to verify the credibility of model outputs. Yet, it is unclear how humans and LLMs evaluate citations when comparing outputs, a process central to reward modeling and modern LLM post-training.
摘要： 许多自然语言处理（NLP）任务要求系统在输出中提供归因，即对基础来源的引用。归因不仅是防止模型产生幻觉的屏障，也是用户验证模型输出可信度的一种手段。然而，目前尚不清楚人类和大型语言模型（LLM）在比较输出时如何评估引用，而这一过程正是奖励建模和现代大模型后训练的核心。
This paper studies the role of citations in the preferences of human judges and four open-source LLMs within the context of scientific question answering, leveraging mixed effects models to investigate the influence of citations on pairwise judgments.
本文在科学问答的背景下，研究了引用在人类评判者和四个开源大模型偏好中的作用，并利用混合效应模型探讨了引用对成对判断的影响。
Among our key findings are (1) that humans prefer more diverse citations but fewer overall, and (2) that LLMs show some citation-related preferences compared to humans, despite lacking access to the sources, but these preferences depend on the data and specific models. We further discuss the implications of our findings for preference data collection.
我们的主要发现包括：（1）人类更倾向于引用来源多样化但引用数量较少的输出；（2）尽管大模型无法直接访问原始来源，但与人类相比，它们仍表现出一定的引用相关偏好，不过这些偏好取决于数据和具体模型。我们进一步讨论了这些发现对偏好数据收集的启示。
Paper Details:
- Authors: Yu Hou, Hal Daumé III, Rachel Rudinger, William Walden
- arXiv ID: 2608.21376
- Subject: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
论文详情：
- 作者： Yu Hou, Hal Daumé III, Rachel Rudinger, William Walden
- arXiv ID: 2608.21376
- 学科分类： 计算与语言 (cs.CL)；人工智能 (cs.AI)