QAngaroo 多步骤推理阅读理解数据集

QAngaroo 数据集是由伦敦大学学院 (UCL) 于 2018 年创建的专注于多跳（multi-hop，也称为多步骤）推理的阅读理解数据集，相关论文成果为「Constructing Datasets for Multi-hop Reading Comprehension Across Documents」。这个数据集包含两个部分：WikiHop 和 MedHop，旨在构建能够执行多跳推理的阅读理解方法，即在不同文档中分散的事实需要通过多个步骤的推理来得出新的事实。

The QAngaroo dataset was developed by University College London (UCL) in 2018. It is a reading comprehension benchmark focused on multi-hop (also referred to as multi-step) reasoning, and its associated academic paper is titled "Constructing Datasets for Multi-hop Reading Comprehension Across Documents". This dataset includes two subsets: WikiHop and MedHop. It aims to facilitate the development of reading comprehension approaches capable of executing multi-hop reasoning, where facts dispersed across different documents require multi-step inference to derive new factual findings.

搜集汇总

背景与挑战

背景概述

QAngaroo是一个专注于多步骤推理的阅读理解数据集，包含WikiHop（开放领域）和MedHop（医学领域）两部分，分别提供数万和数千个训练样本。该数据集旨在支持开发能够进行复杂多跳推理的阅读理解模型。