2WikiMultiHopQA
收藏资源简介:
2WikiMultiHopQA是一个大规模、高质量的多跳问答数据集,由日本的高级研究大学院大学和国家信息学研究所创建。该数据集包含192,606条数据,旨在通过要求模型阅读多个段落来测试推理和推断技能。数据集通过结合结构化和非结构化数据,引入了证据信息,包含多跳问题的推理路径,有助于解释预测和评估模型的推理能力。数据集创建过程中,设计了详细的流程和模板,确保问题的多跳步骤和质量。应用领域包括机器阅读理解,旨在解决需要复杂推理的问答问题。
2WikiMultiHopQA is a large-scale, high-quality multi-hop question answering dataset created by the Graduate University for Advanced Studies and the National Institute of Informatics in Japan. The dataset contains 192,606 instances, and is designed to test models' reasoning and inferential skills by requiring them to read multiple paragraphs. By combining structured and unstructured data, the dataset introduces evidence information and includes the reasoning paths of multi-hop questions, which helps explain model predictions and evaluate the reasoning capabilities of models. During the dataset construction process, detailed workflows and templates were developed to ensure the multi-hop structure and quality of the questions. Its application fields include machine reading comprehension, targeting question answering tasks that require complex reasoning.




