PAQ (Probably Asked Questions)

可能问的问题 (PAQ) 是一个非常大的资源，包含 6500 万个自动生成的 QA 对。 PAQ 是 65M 自然语言 QA 对的半结构化知识库 (KB)，模型可以记忆和/或学习从中检索。 PAQ 与传统 KB 的不同之处在于问题和答案以自然语言存储，并且生成的问题很可能出现在 ODQA 数据集中。 PAQ 是使用问题生成模型和维基百科自动构建的。Potentially Asked Questions (PAQ) is an extremely large-scale resource containing 65 million automatically generated QA pairs. PAQ is a semi-structured knowledge base (KB) composed of 65M natural language QA pairs, from which models can either memorize relevant information or learn to retrieve content. The core distinction between PAQ and traditional KBs is that both questions and answers are stored in natural language, and the generated questions are highly likely to appear in ODQA datasets. PAQ is automatically constructed using question generation models and Wikipedia.

搜集汇总

背景与挑战

背景概述

PAQ (Probably Asked Questions) 是一个包含6500万个自动生成的问答对的大型半结构化知识库，采用自然语言存储问题与答案。该数据集通过问题生成模型和维基百科自动构建，旨在支持开放领域问答任务。