seaoven/fine-tuned-partages-de-identification

French medical de-identification NER model

This model is a CamemBERT-based token classification checkpoint fine-tuned for named entity recognition on French medical text, with a focus on de-identification entities. NER labels: Nom, Prenom, Url, Ville, DateIdentifiante, SituationFamiliale, IdentiteSocialePatient, DateDesidentifiante, Pays, DateNaissancePatient, Adresse, NumeroTelephone, NationalitePatient

Results of training per epoch

Epoch	Training Loss	Validation Loss	Precision	Recall	F1	Macro F1	Accuracy
1	3.259739	3.239809	0.008624	0.192094	0.016506	0.016309	0.234206
2	3.155496	3.177056	0.022145	0.451456	0.042219	0.043363	0.343179
3	3.062432	3.103942	0.026669	0.508322	0.050679	0.069548	0.403236
4	2.965864	3.031749	0.031932	0.550624	0.060364	0.107945	0.472002
5	2.878689	2.966769	0.038775	0.561720	0.072542	0.141406	0.559576
6	2.799171	2.908879	0.041481	0.555479	0.077197	0.138295	0.591547
7	2.737990	2.857519	0.048666	0.576976	0.089762	0.148555	0.644660
8	2.676395	2.815246	0.060018	0.594313	0.109026	0.164073	0.705796
9	2.638831	2.777210	0.059892	0.567268	0.108344	0.168802	0.714798
10	2.573915	2.748089	0.067711	0.587379	0.121425	0.175791	0.739021
11	2.542103	2.719197	0.073217	0.590153	0.130272	0.188839	0.756595
12	2.527237	2.700605	0.073934	0.596394	0.131559	0.189141	0.756458
13	2.507565	2.686029	0.076418	0.597781	0.135513	0.194222	0.764205
14	2.482355	2.677717	0.076399	0.591540	0.135322	0.188957	0.765304
15	2.477236	2.674951	0.076207	0.590153	0.134983	0.189207	0.764813