WikiWiki

该数据集名为WikiWiki，包含了来自1000万篇维基百科文章中的实体和段落，这些实体和段落与包含4.1万个类型的Wikidata知识图谱相连。它被用作细粒度实体分类的基准。此外，WikiWiki在类型精确度方面经过了人工评委的评估，确认其准确率达到85%，相较于其他基准，它具有高质量的数据。该数据集的规模包括1000万篇文章和4.1万个类型，其任务是细粒度实体分类。The dataset named WikiWiki contains entities and paragraphs sourced from 10 million Wikipedia articles, which are linked to the Wikidata knowledge graph encompassing 41,000 entity types. It serves as a benchmark for fine-grained entity classification. Moreover, WikiWiki underwent human evaluation for type precision, with its overall accuracy verified to reach 85%, positioning it as a high-quality dataset relative to other existing benchmarks. With a corpus scale of 10 million articles and 41,000 entity types, this dataset is tailored for the fine-grained entity classification task.

搜集汇总

背景与挑战

背景概述

WikiWiki是一个大规模实体类型标注数据集，从1000万维基百科文档中提取了约220万实体、4万实体类型和3900万提及的三元组数据。该数据集包含上下文段落、提及到实体的映射和实体到类型的映射，主要用于链接、实体类型标注任务，并支持通过多任务QA将类型知识注入语言模型。