WD-KNOWN
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/bveseli/LMsForKBC
下载链接
链接失效反馈官方服务：
资源简介：
该数据集名为WD-KNOWN，是从维基数据中抽取的公正随机样本，包含了超过390万条事实，旨在对语言模型的知识库补全能力进行真实评估。该数据集不仅能够测试语言模型预测多标记对象的能力，而且还包含了一个大规模样本，该样本不受限于拥有维基百科页面的主体。总体而言，该数据集涵盖了超过390万条事实，任务集中在知识库补全（KBC）方面。
The dataset named WD-KNOWN is an unbiased random sample extracted from Wikidata, comprising over 3.9 million facts. It is designed to conduct authentic evaluations of the knowledge base completion capabilities of language models. This dataset not only can test the ability of language models to predict multi-label objects, but also includes a large-scale sample that is not restricted to entities with Wikipedia pages. Overall, this dataset encompasses over 3.9 million facts, with its tasks focused on Knowledge Base Completion (KBC).
提供机构：
Authors of the paper
