CODEX
收藏资源简介:
CODEX数据集是由密歇根大学的Tara Safavi和Danai Koutra开发的,旨在改进现有的知识图谱完成基准。该数据集从Wikidata和Wikipedia提取,包含三个不同大小和结构的知识图谱,涵盖多语言实体和关系描述,以及数以万计的硬负例。CODEX数据集特别适用于评估模型在处理多样的、可解释的内容和更复杂的链接预测任务上的表现。此外,数据集还包含了详细的逻辑关系模式分析和基准测试实验,以支持知识图谱完成方法的进一步发展和评估。
The CODEX dataset was developed by Tara Safavi and Danai Koutra from the University of Michigan to advance existing knowledge graph completion benchmarks. Extracted from Wikidata and Wikipedia, this dataset includes three knowledge graphs with distinct sizes and structures, covering multilingual entity and relation descriptions as well as tens of thousands of hard negative samples. The CODEX dataset is specifically designed to evaluate model performance on handling diverse, interpretable content and more complex link prediction tasks. Furthermore, the dataset contains detailed logical relation pattern analyses and benchmark experiments to support the further development and evaluation of knowledge graph completion methods.

- 1CoDEx: A Comprehensive Knowledge Graph Completion Benchmark密歇根大学 · 2020年



