search

JULIELab / trec-pm

Support code and resources for participation at the TREC Precision Medicine Track (TREC-PM)
http://trec-cds.appspot.com
MIT License
9 stars 2 forks source link

Create manual curated filter list for treatments #77

Closed khituras closed 5 years ago

khituras commented 5 years ago

This concerns e.g. #76. Some treatment words are "various" and other probably non-helpful stuff. Those should be filtered out. The easiest way would be to remove such words from the index directly. When creating filter lists, check out if the broadTreatments field is mainly containing the bad candidates and whether focusedTreatments fares significantly better. Also check whether the focused field is missing good treatment concepts which occur in the broad field.

michelole commented 5 years ago

Blocked on #41 (and somehow, #42).

michelole commented 5 years ago

Check whether the broadTreatments contain e.g. surgery and radiotherapy, which are valid treatments currently not found with the focusedTreatments.

michelole commented 5 years ago

Fixed by cf97263800cdb4aa6b9518dedf48facd424e9850. broadTreatments seem to only add additional noise with no benefit. I'm happy with the filtered focusedTreatments so far.