Pretrained model for entity linking

[dzieciou]

Feature description

Is there a pretrained model for Entity Linking using Wikipedia/WikiData as DB?

Spacy recently started to support Entity Linking 1,2,3 but they do not provide pretrained model for that and some people report training takes at least a week 4.

Could the feature be a custom component or spaCy plugin?

If so, we will tag it as project idea so other users can take it on.

I guess it should a separate model file.

[svlandeg]

No, that's currently not readily available, the feature is still in beta development. You can use the example scripts provided to train your own, on Wikipedia/Wikidata or any other resource. Obviously it depends on which data you train it on, and how much of it, how long the training will take. I don't think it is not necessary for most use-cases to train it on the whole of English Wikipedia, but that's your choice. Assembling the KB and training on 165.000 articles, took me a few days total.

See also the discussion here: #4511