Permalink
Switch branches/tags
Nothing to show
Find file Copy path
315a4ec May 12, 2017
1 contributor

Users who have contributed to this file

15 lines (8 sloc) 979 Bytes

Performance of trained models

Dataset State-of-the-art F1-score NeuroNER F1-score Model Entities
CoNLL-2003-en 90.9 90.5 Link Location, misc, organization, person
i2b2 2014 97.9* 97.7* Link 18 PHI types
MIMIC deid 2016 98.5* 98.6* Link 18 PHI types

PHI = protected health information.

* indicates that the F1-score is binary, i.e. that the predicted class of the named-entity is not taken into account when computing the F1-score. This is most important metric for some use cases such as de-identification, where the main goal is to tag as many PHI (protected health information) entities as possible, the correct prediction of the PHI class being secondary.