CRaQAn (Coreference Resolution in Question-Answering)

CRaQAn数据集是由边缘分析公司和Pyxos公司联合开发的，专注于问答任务中的指代消解问题。该数据集包含261个经过人工审核的问答样本，每个样本都涉及跨句子的指代消解。数据集的内容主要来源于维基百科关于现代美国法律的文章，这些文章因其复杂的指代关系而被选中。创建过程中，使用了GPT-4和递归批评与改进循环（RCI）技术来自动生成高质量的数据集。CRaQAn数据集的应用领域主要集中在测试和评估问答系统中的信息检索策略，特别是在处理长文档时的指代消解能力。

The CRaQAn dataset was co-developed by Edge Analytics and Pyxos, focusing on coreference resolution in question answering (QA) tasks. This dataset contains 261 manually reviewed QA samples, each involving cross-sentence coreference resolution. The content of the dataset is primarily sourced from Wikipedia articles on modern American law, which were selected for their complex coreference relations. During the dataset's development, GPT-4 and the Recursive Criticism and Improvement (RCI) technique were utilized to automatically generate high-quality samples for this dataset. The main application areas of the CRaQAn dataset focus on testing and evaluating information retrieval strategies in QA systems, particularly the coreference resolution capability when processing long documents.