feat(phase2): Fine-tuning CamemBERT-bio v2 (F1=0.90) + enrichissement données

Browse Source

- Fine-tuning camembert-bio-base : F1=0.903, Recall=0.930 (vs 0.89/0.85)
- Data augmentation : substitution noms INSEE (219K patronymes, x3 copies)
- Hard negatives BDPM (5.7K médicaments) + QUAERO (1319 termes médicaux)
- Annotations silver enrichies par gazetteers (+612 VILLE, +5 HOPITAL)
- Export silver avec support multi-répertoires (--extra-dir)
- Gazetteers QUAERO : CHEM, DISO, PROC, ANAT depuis DrBenchmark/QUAERO
- Gazetteers INSEE : noms de famille fréquents (96K) et complets (219K)
- Batch silver 1194 PDFs (run_batch_silver_export.py) pour dataset v3

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

...

This commit is contained in:

Domi31tls

2026-03-10 02:06:08 +01:00

parent 274e2fa586

commit c9572c383a

38 changed files with 318811 additions and 1406 deletions

Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL

Download Patch File Download Diff File Expand all files Collapse all files

218982

data/insee/noms_famille_france.txt Normal file

File diff suppressed because it is too large Load Diff

96198

data/insee/noms_famille_frequents.txt Normal file

File diff suppressed because it is too large Load Diff