HAGRID

HAGRID是由滑铁卢大学计算机科学学院创建的一个用于生成信息检索模型的数据集，包含2638个查询。该数据集通过人类与大型语言模型（如GPT-3.5）的协作构建，旨在通过引用支持来源来生成解释性答案。数据集内容包括信息检索问题及其相关文本片段，以及由GPT-3.5生成的答案。创建过程中，首先由语言模型自动收集引用解释，然后由人类注释者评估其信息性和可归属性。HAGRID的应用领域主要集中在开发具有更好引用能力的信息检索模型，以解决模型输出缺乏足够知识来源支持的问题。

HAGRID is a dataset for developing information retrieval models, created by the School of Computer Science at the University of Waterloo, which contains 2638 queries. This dataset is built through the collaboration between human annotators and large language models (LLMs, e.g., GPT-3.5), with the objective of generating explanatory answers supported by cited sources. The dataset encompasses information retrieval questions, their associated text snippets, as well as answers generated by GPT-3.5. During the construction process, language models were first utilized to automatically collect cited explanations, which were then evaluated for informativeness and attributability by human annotators. The primary applications of HAGRID focus on developing information retrieval models with enhanced citation capabilities, to address the issue where model outputs lack sufficient support from valid knowledge sources.