Wikidata5m

Info

Wikidata5m is a million-scale knowledge graph dataset with aligned corpus. This dataset integrates the Wikidata knowledge graph and Wikipedia pages. Each entity in Wikidata5m is described by a corresponding Wikipedia page, which enables the evaluation of link prediction over unseen entities.

The dataset is distributed as a knowledge graph, a corpus, and aliases. We provide both transductive and inductive data splits used in the original paper.

Data

• Knowledge graph: Transductive split, 160 MB. Inductive split, 160 MB. Raw, 168 MB.
• Corpus, 991 MB.
• Entity & relation aliases, 188 MB.

For raw knowledge graph, it may also contain entities that do not have corresponding Wikipedia pages.

Format

Wikidata5m follows the identifier system used in Wikidata. Each entity and relation is identified by a unique ID. Entities are prefixed by Q, while relations are prefixed by P.

Knowledge Graph

The knowledge graph is stored in the triplet list format. For example, the following line corresponds to <Donald Trump, position held, President of the United States>.

Q22686	P39	Q11696

Corpus

Each line in the corpus is a document, indexed by entity ID. The following line shows the description for Donald Trump.

Q22686	Donald John Trump (born June 14, 1946) is the 45th and current president of the United States ...

Alias

Each line lists the alias for an entity or relation. The following line shows the aliases of Donald Trump.

Q22686  donnie trump	45th president of the united states     Donald John Trump ...

Publications

• KEPLER: A Unified Model for Knowledge Embedding and Pre-trained Language Representation
  Xiaozhi Wang, Tianyu Gao, Zhaocheng Zhu, Zhengyan Zhag, Zhiyuan Liu, Juanzi Li, Jian Tang
  TACL 2021 arXiv BibTeX