Stephan Tulkens, Machine Learning Engineer at Slimmer AI, recently had a case study in Towards Data Science called, “Doing almost as much with much less: A Case Study in Biomedical Named Entity Recognition”.
At the beginning of any new R&D project, I look to match the most appropriate solution to the problem at hand — bringing the solution into production without compromising on the outcomes I’d like to support.
In his investigation, Stephan used a string-matching system, QuickUMLS, as a classifier for the MedMentions dataset. His goal was to see if it could compete with a supervised model on a biomedical Named Entity Recognition (BioNER) task. In the absence of vast amounts of training data, this approach yielded robust results and is a simple way to implement in product.