This is a dataset for the training of a type classifier on Wikidata entity types. It is derived from Wikidata-Disamb by Cetoli et el. (2019).
The train set has 100,000 samples. The test and the dev set have 10,000 samples each. Each sample has four values:
- A natural language sentence s
- A mention m (referring to an entity e)
- The entity e (to which the mention m refers)
- The entity type t of the entity e
There are three versions for each part of the dataset:
- small: small part of the dataset for debugging
- medium: larger part of the dateset for tests and debugging
- full: full dataset
The types are a predefined set derived from Wikidata Concept Monitor taxonomy:
| Wikidata item | Type |
|---|---|
| http://www.wikidata.org/entity/Q215627 | person |
| http://www.wikidata.org/entity/Q163875 | cardinal |
| http://www.wikidata.org/entity/Q838948 | work of art |
| http://www.wikidata.org/entity/Q13442814 | article in scholarly journal |
| http://www.wikidata.org/entity/Q571 | book |
| http://www.wikidata.org/entity/Q618123 | geographical feature |
| http://www.wikidata.org/entity/Q43229 | organization |
| http://www.wikidata.org/entity/Q811979 | architectural structure |
| http://www.wikidata.org/entity/Q16521 | taxon |
| http://www.wikidata.org/entity/Q1656682 | event |
| http://www.wikidata.org/entity/Q83620 | thoroughfare |
| other | other |