prevent subword annotations using existing medacy models #61

New issue

Open

opened 2025-10-14 16:21:53 -06:00 by navan · 0 comments

navan commented

2025-10-14 16:21:53 -06:00

Owner

Originally created by @eniale88 on 10/17/2019

I tested medacy_model_clinical_notes on clinical reports, and noticed that it produces some false positives when tagging on subword level.

For instance,
model.predict("Triggering abnormal complications")

results in

{'entities': {'T1': ('Drug', 11, 13, 'ab')}, 'relations': []}

How can subword-level tagging be prevented when using a medacy model?

*Originally created by @eniale88 on 10/17/2019* I tested **medacy_model_clinical_notes** on clinical reports, and noticed that it produces some false positives when tagging on subword level. For instance, `model.predict("Triggering abnormal complications")` results in `{'entities': {'T1': ('Drug', 11, 13, 'ab')}, 'relations': []}` How can subword-level tagging be prevented when using a medacy model?