PlausibleQA
收藏资源简介:
PlausibleQA是一个大规模的问题回答数据集,由因斯布鲁克大学创建。该数据集包含10,000个问题及每个问题的100个候选答案,每个答案都标注了可信度分数和选择理由。此外,数据集还提供了90万个候选答案之间的两两比较理由,用于进一步细化可信度评估。该数据集旨在为问答系统的研究和大型语言模型性能的提升提供资源,特别是在区分可信的干扰项和正确答案方面。
PlausibleQA is a large-scale question answering dataset created by the University of Innsbruck. It contains 10,000 questions along with 100 candidate answers for each question, where each answer is annotated with a credibility score and a justification for its selection. Additionally, the dataset provides pairwise comparison justifications between 900,000 candidate answers to further refine credibility assessments. This dataset aims to provide resources for research on question answering systems and the improvement of large language model performance, particularly in distinguishing between plausible distractors and correct answers.




