Natural Questions
收藏资源简介：
自然问题语料库是一个问答数据集，包含 307,373 个训练示例、7,830 个开发示例和 7,842 个测试示例。每个示例都由 google.com 查询和相应的 Wikipedia 页面组成。每个 Wikipedia 页面都有一个在回答问题的页面上注释的段落（或长答案），以及包含实际答案的注释段落的一个或多个短跨度。然而，长答案和短答案注释可以是空的。如果它们都是空的，那么页面上根本没有答案。如果长答案注释不为空，而短答案注释为空，则注释的段落回答了问题，但找不到明确的简短答案。最后，有 1% 的文档有一段用“是”或“否”的简短答案注释的段落，而不是短跨度列表。
The Natural Questions Corpus is a question answering dataset containing 307,373 training examples, 7,830 development examples, and 7,842 test examples. Each example consists of a Google.com query and its corresponding Wikipedia page. For each Wikipedia page, there is a paragraph (or long answer) annotated on the page that answers the question, as well as one or more short spans from the annotated paragraph that contain the actual answer. However, both the long answer and short answer annotations can be empty. If both are empty, then there is no answer at all on the page. If the long answer annotation is non-empty but the short answer annotation is empty, the annotated paragraph answers the question but no explicit short answer can be found. Finally, 1% of the documents have a paragraph annotated with a "yes" or "no" short answer instead of a list of short spans.
- Natural Questions数据集首次由Google Research团队在ACL会议上发表，旨在为开放域问答系统提供一个标准化的评估基准。
- 该数据集被广泛应用于多个问答系统的研究和开发中，成为评估模型性能的重要工具。
- 随着自然语言处理技术的进步，Natural Questions数据集开始用于探索更复杂的问答任务，如多跳问答和长答案生成。
