Dom/anonymisation
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: téléphone +33(0) non détecté + noms médecins homonymes de termes médicaux

Browse Source 
- RE_TEL : ajout du format +33(0)XXXXXXXXX (ex: +33(0)156125400)
- _add_tokens_force_first : tous les tokens après Dr/Mme/Mr sont maintenant
  dans force_names (bypass stop-words médicaux). Corrige la fuite de noms
  de médecins homonymes de termes médicaux (ex: Dr MASSE)

Score évaluation maintenu à 100.0/100 (A+)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Domi31tls
2026-03-16 18:33:32 +01:00
parent 27d19ebed7
commit 7c05ff9aaf
1 changed files with 9 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
anonymizer_core_refactored_onnx.py
View File

@@ -372,7 +372,7 @@ CRITICAL_PII_KEYS = {"EMAIL", "TEL", "IBAN", "NIR", "IPP", "DATE_NAISSANCE"}
# Baseline regex # Baseline regex
RE_EMAIL = re.compile(r"[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}") RE_EMAIL = re.compile(r"[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}")
RE_URL = re.compile(r"https?://[A-Za-z0-9._~:/?#\[\]@!$&'()*+,;=\-%]+", re.IGNORECASE) RE_URL = re.compile(r"https?://[A-Za-z0-9._~:/?#\[\]@!$&'()*+,;=\-%]+", re.IGNORECASE)
RE_TEL = re.compile(r"(?<!\d)(?:\+33\s?|0)\d(?:[\s.\-]?\d){8}(?!\d)") RE_TEL = re.compile(r"(?<!\d)(?:\+33\s?(?:\(0\))?\s?|0)\d(?:[\s.\-]?\d){8}(?!\d)")
RE_TEL_COMPACT = re.compile(r"(?<!\d)0[1-9]\d{8}(?!\d)") RE_TEL_COMPACT = re.compile(r"(?<!\d)0[1-9]\d{8}(?!\d)")
RE_IBAN = re.compile(r"\b[A-Z]{2}\d{2}(?:\s?[A-Z0-9]{4}){3,7}(?:\s?[A-Z0-9]{1,4})\b") RE_IBAN = re.compile(r"\b[A-Z]{2}\d{2}(?:\s?[A-Z0-9]{4}){3,7}(?:\s?[A-Z0-9]{1,4})\b")
RE_IPP = re.compile(r"\b(?:I\.?P\.?P\.?|IPP|N°\s*Ipp)\s*[:\-]?\s*([A-Za-z0-9]{6,})\b", re.IGNORECASE) RE_IPP = re.compile(r"\b(?:I\.?P\.?P\.?|IPP|N°\s*Ipp)\s*[:\-]?\s*([A-Za-z0-9]{6,})\b", re.IGNORECASE)
@@ -1983,24 +1983,21 @@ def _extract_document_names(full_text: str, cfg: Dict[str, Any]) -> Tuple[set, s
force_names.add(token) force_names.add(token)
def _add_tokens_force_first(match_str): def _add_tokens_force_first(match_str):
"""Comme _add_tokens mais force le 1er token (contexte Dr/Mme fort).""" """Comme _add_tokens mais force TOUS les tokens (contexte Dr/Mme fort).
Après Dr/Mme, tous les tokens sont des noms — même s'ils sont
homonymes de termes médicaux (ex: Dr Laurence MASSE).
"""
_add_compound(match_str) _add_compound(match_str)
tokens = match_str.split() tokens = match_str.split()
for i, token in enumerate(tokens): for token in tokens:
token = token.strip(" .-'") token = token.strip(" .-'")
if len(token) < 4: if len(token) < 4:
continue continue
if token.upper() in wl_sections or token in wl_phrases: if token.upper() in wl_sections or token in wl_phrases:
continue continue
if token.lower() in _MEDICAL_STOP_WORDS_SET:
continue
if i == 0:
# Premier token après Dr/Mme : contexte fiable
names.add(token)
else:
if len(token) < 4:
continue
names.add(token) names.add(token)
force_names.add(token)
for m in RE_EXTRACT_PATIENT.finditer(full_text): for m in RE_EXTRACT_PATIENT.finditer(full_text):
_add_tokens_force_all(m.group(1)) _add_tokens_force_all(m.group(1))