Entity linking is the task of detecting mentions of entities from a knowledge base in a document.

Are there any openly available datasets to train and evaluate such systems? The datasets I came across are all closed (for instance, TAC-KBP is owned by the Linguistic Data Consortium, the YAGO-AIDA dataset requires access to the original CoNLL 2003 dataset, and so on).

The dataset should contain some text with reference annotations (for instance links to Wikipedia), and should be of reasonable size for machine learning purposes. For instance, Wikipedia dumps themselves provide this to some extent, but the wikilinks are generally not intended to be complete, so the annotations are very sparse.


The IITB dataset is quite small compared to the TAC datasets (107 documents) but it is freely available and has higher recall than off-the-shelf Wikipedia pages. It can be downloaded at https://www.cse.iitb.ac.in/~soumen/doc/CSAW/Annot/

| improve this answer | |
  • The RDF2NLP project also lists some datasets in the NIF format - not all of them are relevant for Entity Linking though: dashboard.nlp2rdf.aksw.org – pintoch Dec 28 '17 at 11:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.