Add support for biomedical NER datasets

[stefan-it]

A great enhancement would be to have built-in dataset loaders for biomedical NER datasets.

A good resource are the datasets mentioned in the BioBERT paper. The datasets can be downloaded form the BioBERT repository.

Thus, the following datasets can be supported:

• NCBI disease
• BC5CDR (Disease and Drug/Chemical)
• BC4CHEMD
• BC2GM
• JNLPBA
• LINNAEUS
• Species-800

[alanakbik]

Yes that's a great idea! Together with the ELMo pubmed model (#502) and the Flair pubmed embeddings (#518) this could enable more research into biomedical data.

[stefan-it]

For now I'm going to write importers for the following datasets:

• JNLPBA
• NCBI-disease
• bc5cdr

Thanks to the SciBERT authors, these datasets are already preprocessed and can simply be fetched from their GitHub repository :)

[shreyashub]

@stefan-it Were you able to write importers for some of the datasets? If yes, could you reference the PR here?
If no, then I could help in that, along with the other medical NER datasets that were not in SciBERT.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[alanakbik]

There is a highly active branch currently in which my colleagues are adding a large amount of different biomedical datasets. See #1513. I think it will be merged soon!

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[alanakbik]

Support for biomedical datasets was added in Flair 0.6. If any datasets are missing let us know!