Documentation for Entity Linking

[svlandeg]

Documentation for PRs #3459, #3864 and PR #4003 - all referring to the Entity Linking functionality discussed in Issue #3339.

Added documentation

API

• NEW: kb.md (for KnowledgeBase in kb.pyx)
• NEW: candidate.md (for Candidate in kb.pyx)
• NEW: entitylinker.md (for EntityLinker in pipes.pyx)
• UPDATED: span.md, added kb_id attribute
• UPDATED: token.md, added ent_kb_id attribute
• UPDATED: goldparse.md, added links attribute
• UPDATED: entityrecognizer.md, textcategorizer.md and tagger.md, with more explicit references to tensors and optimizer.averages (to keep the docs consistent across pipe components)

Usage

• UPDATED: 101/_named_entities.md and 101/_pipelines.md, added explanations around KB ID fields
• UPDATED: processing-pipelines.md, added entity_linker pipe explanation
• UPDATED: spacy-101.md, added Entity Linking & Knowledge Base definitions
• UPDATED: training.md with information on training the entity linker + references to the two generic example scripts
• UPDATED: linguistic-features.md, added explanations around Entity Linking + example code
• UPDATED: facts-figures.md, set support to Entity Linking to True !

TODO before merging

• ctrl-F "X.X" and replace with actual version number. DONE
• Make sure the pretrained models include the entity_linking pipe so the code snippets actually work. Some of them currently don't, I tested them on custom models instead. The example scripts should work though because they start with a blank model.
• Update language pipe overview picture and add EL functionality ?

Types of change

Changes to the documentation

Checklist

• I have submitted the spaCy Contributor Agreement.
• I ran the tests, and all new and existing tests passed.
• My changes don't require a change to the documentation, or if they do, I've added all required information.

[svlandeg]

Update: filtered this PR down and moved all functionality changes to PR #4091 --> once the latter one is merged, will merge into this to clean this up.
[EDIT: done]

[svlandeg]

@ines : thanks for the review! I think I addressed everything. I was also wondering whether I should add anything to the Cython documentation (strucs & classes), or do we consider the KB too specific to be included there?

[ines]

I was also wondering whether I should add anything to the Cython documentation (strucs & classes), or do we consider the KB too specific to be included there?

I think for now, it's okay to not document the Cython part and consider this internals. There's a chance that some of the specifics are still going to change. The main focus of the Cython docs are the data structures that we want users to interact with directly from their Cython implementations.