HoVer

是用于多跳证据提取和事实验证的数据集。它挑战模型以从与声明相关的几篇 Wikipedia 文章中提取事实，并分类声明是否被事实支持或不支持。在 HoVer 中，索赔要求从多达四篇英文维基百科文章中提取证据，并体现各种形状的推理图。

This is a dataset for multi-hop evidence extraction and fact verification. It tasks models with extracting factual evidence from multiple Wikipedia articles relevant to a given claim, and classifying whether the claim is supported or refuted by the evidence. In the HoVer dataset, each claim requires evidence extracted from up to four English Wikipedia articles, and encompasses reasoning graphs of various shapes.

搜集汇总

背景与挑战

背景概述

HoVer是一个用于多跳证据提取和事实验证的数据集，要求模型从多篇英文维基百科文章中提取事实并分类声明的支持性。该数据集发布于2020年，适用于文本自然语言处理和多跳问答等任务。