This is the repo accompanying the blog post "Doing almost as much with much less: a case study in biomedical named entity recognition". In short, we use QuickUMLS to classify ambiguous named entities from text into a pre-specified set of semantic types. While the baseline performance of QuickUMLS is pretty good, we show that it can be a lot better with some extra tweaks.

For the full information, please take a look at the blog post here.

Install all the requirements in requirements.txt. The version numbers for numpy and spacy have been omitted. This code should work with either spacy 2 or 3.

Then, using QuickUMLS, you should create a QuickUMLS installation somewhere. For instructions on how to do that, see here. Once you've done this, you're all set.

To reproduce our results, you can download the Medmentions corpus here. If you put the MedMentions corpus files in data/, the file run.py should run out of the box.

Stéphan Tulkens

MIT