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.