ASNQ

该数据集名为ASNQ，是针对答案句子选择（AS2）任务的传统基准数据集，其问题来源于必应查询日志，而答案则从用户点击的维基百科页面中提取。问题可能仅包含正确答案，仅包含错误答案，或者正确与错误答案混合，这种情况会影响它们在学习重排器时的实用性。该数据集的规模包括1229个训练问题、65个验证问题和68个测试问题，分别对应53,417、1,117和1,442个问题-答案对。所涉及的任务是答案句子选择（AS2）。

This dataset, named ASNQ, is a traditional benchmark dataset for the Answer Sentence Selection (AS2) task. Its questions are sourced from Bing search query logs, and the answers are extracted from Wikipedia pages that users clicked on. Questions may contain only correct answers, only incorrect answers, or a mix of both, which undermines their utility when learning rerankers. The dataset includes 1,229 training questions, 65 validation questions, and 68 test questions, corresponding to 53,417, 1,117, and 1,442 question-answer pairs respectively. The task involved is Answer Sentence Selection (AS2).