PAT-Questions
收藏资源简介:
PAT-Questions是由加州大学河滨分校创建的一个包含6172个问题-答案对的数据集,专注于当前时间锚定的时态问答(PATQA)。数据集内容涵盖单跳和多跳时态问题,通过SPARQL查询自动更新答案,确保数据的时效性。创建过程利用Wikidata知识库,通过模板生成问题,并关联每个问题的SPARQL查询以准确检索最新答案。该数据集旨在解决大型语言模型在处理时态问题时的知识更新和复杂时态关系推理的挑战,适用于测试和提升模型在实时信息处理和多步时态推理方面的能力。
PAT-Questions is a dataset comprising 6,172 question-answer pairs, developed by the University of California, Riverside, and focused on current-time-anchored temporal question answering (PATQA). The dataset includes both single-hop and multi-hop temporal questions, with answers automatically updated via SPARQL queries to ensure data timeliness. Its construction leverages the Wikidata knowledge base, generating questions through templates and linking each question to a corresponding SPARQL query to accurately retrieve the latest answers. This dataset aims to address the challenges of knowledge updating and complex temporal relation reasoning faced by large language models (LLMs) when processing temporal questions, and is designed to test and enhance models' capabilities in real-time information processing and multi-step temporal reasoning.




