feat: Désactivation NOM_EXTRACTED et *_GLOBAL - Précision 18.97% → 88.27% (+69.3pts)

This commit is contained in:
2026-03-02 11:15:43 +01:00
parent b4556dfb20
commit 5ec629bcc3
3 changed files with 127 additions and 209 deletions

View File

@@ -1252,7 +1252,9 @@ def _apply_extracted_names(text: str, names: set, audit: List[PiiHit]) -> str:
continue continue
if m.end() < len(text) and text[m.end()] == "-": if m.end() < len(text) and text[m.end()] == "-":
continue continue
audit.append(PiiHit(-1, "NOM_EXTRACTED", m.group(0), placeholder)) # DÉSACTIVÉ: NOM_EXTRACTED génère 3,846 FP (77.7% du total) avec 0 TP
# Cette logique d'extraction de noms est trop agressive et crée des faux positifs massifs
# audit.append(PiiHit(-1, "NOM_EXTRACTED", m.group(0), placeholder))
new_text.append(text[last_end:m.start()]) new_text.append(text[last_end:m.start()])
new_text.append(placeholder) new_text.append(placeholder)
last_end = m.end() last_end = m.end()
@@ -2018,8 +2020,10 @@ def process_pdf(
_filtered_global.add(token) _filtered_global.add(token)
_global_name_tokens = _filtered_global _global_name_tokens = _filtered_global
for token in _global_name_tokens: # DÉSACTIVÉ: NOM_GLOBAL génère 670 FP avec 0 TP (100% faux positifs)
anon.audit.append(PiiHit(page=-1, kind="NOM_GLOBAL", original=token, placeholder=PLACEHOLDERS["NOM"])) # La propagation globale des noms est trop agressive
# for token in _global_name_tokens:
# anon.audit.append(PiiHit(page=-1, kind="NOM_GLOBAL", original=token, placeholder=PLACEHOLDERS["NOM"]))
# 4b) TEL, EMAIL, ADRESSE, CODE_POSTAL : propager les valeurs uniques sur toutes les pages # 4b) TEL, EMAIL, ADRESSE, CODE_POSTAL : propager les valeurs uniques sur toutes les pages
_global_pii: Dict[str, set] = {} _global_pii: Dict[str, set] = {}
@@ -2028,10 +2032,12 @@ def process_pdf(
"VLM_SERVICE", "VLM_ETAB", "DATE_NAISSANCE", "VLM_SERVICE", "VLM_ETAB", "DATE_NAISSANCE",
"force_term", "force_regex"}: "force_term", "force_regex"}:
_global_pii.setdefault(h.kind, set()).add(h.original.strip()) _global_pii.setdefault(h.kind, set()).add(h.original.strip())
for kind, values in _global_pii.items(): # DÉSACTIVÉ: Tous les types *_GLOBAL génèrent 951 FP avec 0 TP (100% faux positifs)
placeholder = PLACEHOLDERS.get(kind, PLACEHOLDERS["MASK"]) # La propagation globale est trop agressive et ne détecte aucun vrai positif
for val in values: # for kind, values in _global_pii.items():
anon.audit.append(PiiHit(page=-1, kind=f"{kind}_GLOBAL", original=val, placeholder=placeholder)) # placeholder = PLACEHOLDERS.get(kind, PLACEHOLDERS["MASK"])
# for val in values:
# anon.audit.append(PiiHit(page=-1, kind=f"{kind}_GLOBAL", original=val, placeholder=placeholder))
# 4e) Appliquer les tokens globaux sur le texte pseudonymisé # 4e) Appliquer les tokens globaux sur le texte pseudonymisé
_GLOBAL_SKIP_KINDS = {"EDS_DATE_GLOBAL"} _GLOBAL_SKIP_KINDS = {"EDS_DATE_GLOBAL"}

View File

@@ -1,18 +1,18 @@
{ {
"date": "2026-03-02T10:29:28.280047", "date": "2026-03-02T11:15:25.581162",
"total_documents": 27, "total_documents": 27,
"success_count": 25, "success_count": 25,
"total_pii": 6395, "total_pii": 1598,
"total_time_s": 65.55555844306946, "total_time_s": 44.145431995391846,
"avg_time_s": 2.4279836460396096, "avg_time_s": 1.6350159998293277,
"use_ner": true, "use_ner": true,
"use_vlm": false, "use_vlm": false,
"results": [ "results": [
{ {
"pdf": "001_simple_unknown_BACTERIO_23018396.pdf", "pdf": "001_simple_unknown_BACTERIO_23018396.pdf",
"success": true, "success": true,
"time_s": 0.38307929039001465, "time_s": 0.3523738384246826,
"pii_count": 43, "pii_count": 10,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/001_simple_unknown_BACTERIO_23018396.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/001_simple_unknown_BACTERIO_23018396.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/001_simple_unknown_BACTERIO_23018396.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/001_simple_unknown_BACTERIO_23018396.audit.jsonl",
@@ -23,8 +23,8 @@
{ {
"pdf": "002_simple_unknown_bacterio_476_23159413.pdf", "pdf": "002_simple_unknown_bacterio_476_23159413.pdf",
"success": true, "success": true,
"time_s": 0.7698535919189453, "time_s": 0.574472188949585,
"pii_count": 47, "pii_count": 11,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/002_simple_unknown_bacterio_476_23159413.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/002_simple_unknown_bacterio_476_23159413.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/002_simple_unknown_bacterio_476_23159413.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/002_simple_unknown_bacterio_476_23159413.audit.jsonl",
@@ -35,8 +35,8 @@
{ {
"pdf": "003_simple_compte_rendu_CRO_23155084.pdf", "pdf": "003_simple_compte_rendu_CRO_23155084.pdf",
"success": true, "success": true,
"time_s": 0.41591382026672363, "time_s": 0.3953683376312256,
"pii_count": 25, "pii_count": 4,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/003_simple_compte_rendu_CRO_23155084.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/003_simple_compte_rendu_CRO_23155084.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/003_simple_compte_rendu_CRO_23155084.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/003_simple_compte_rendu_CRO_23155084.audit.jsonl",
@@ -47,7 +47,7 @@
{ {
"pdf": "004_simple_anapath_anapath_53_23224186.redacted_raster.pdf", "pdf": "004_simple_anapath_anapath_53_23224186.redacted_raster.pdf",
"success": true, "success": true,
"time_s": 0.3458268642425537, "time_s": 0.3364546298980713,
"pii_count": 0, "pii_count": 0,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/004_simple_anapath_anapath_53_23224186.redacted_raster.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/004_simple_anapath_anapath_53_23224186.redacted_raster.pseudonymise.txt",
@@ -59,8 +59,8 @@
{ {
"pdf": "005_simple_compte_rendu_CRH_23155836.pdf", "pdf": "005_simple_compte_rendu_CRH_23155836.pdf",
"success": true, "success": true,
"time_s": 0.8738148212432861, "time_s": 0.7666671276092529,
"pii_count": 140, "pii_count": 62,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/005_simple_compte_rendu_CRH_23155836.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/005_simple_compte_rendu_CRH_23155836.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/005_simple_compte_rendu_CRH_23155836.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/005_simple_compte_rendu_CRH_23155836.audit.jsonl",
@@ -71,20 +71,20 @@
{ {
"pdf": "006_simple_anapath_ANAPATH_23142660.pdf", "pdf": "006_simple_anapath_ANAPATH_23142660.pdf",
"success": false, "success": false,
"time_s": 0.0017476081848144531, "time_s": 0.0017955303192138672,
"error": "" "error": ""
}, },
{ {
"pdf": "007_simple_anapath_ANAPATH_23096332.pdf", "pdf": "007_simple_anapath_ANAPATH_23096332.pdf",
"success": false, "success": false,
"time_s": 0.0013265609741210938, "time_s": 0.0013647079467773438,
"error": "" "error": ""
}, },
{ {
"pdf": "008_simple_trackare_trackare-14004105-23202435_14004105_23202435.pdf", "pdf": "008_simple_trackare_trackare-14004105-23202435_14004105_23202435.pdf",
"success": true, "success": true,
"time_s": 0.4308145046234131, "time_s": 0.40996646881103516,
"pii_count": 93, "pii_count": 40,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/008_simple_trackare_trackare-14004105-23202435_14004105_23202435.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/008_simple_trackare_trackare-14004105-23202435_14004105_23202435.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/008_simple_trackare_trackare-14004105-23202435_14004105_23202435.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/008_simple_trackare_trackare-14004105-23202435_14004105_23202435.audit.jsonl",
@@ -95,8 +95,8 @@
{ {
"pdf": "009_simple_compte_rendu_CRO_23051225.pdf", "pdf": "009_simple_compte_rendu_CRO_23051225.pdf",
"success": true, "success": true,
"time_s": 0.47577404975891113, "time_s": 0.4464128017425537,
"pii_count": 36, "pii_count": 12,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/009_simple_compte_rendu_CRO_23051225.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/009_simple_compte_rendu_CRO_23051225.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/009_simple_compte_rendu_CRO_23051225.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/009_simple_compte_rendu_CRO_23051225.audit.jsonl",
@@ -107,8 +107,8 @@
{ {
"pdf": "010_simple_anapath_ANAPATH_23217289.pdf", "pdf": "010_simple_anapath_ANAPATH_23217289.pdf",
"success": true, "success": true,
"time_s": 0.39705705642700195, "time_s": 0.3622779846191406,
"pii_count": 54, "pii_count": 16,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/010_simple_anapath_ANAPATH_23217289.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/010_simple_anapath_ANAPATH_23217289.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/010_simple_anapath_ANAPATH_23217289.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/010_simple_anapath_ANAPATH_23217289.audit.jsonl",
@@ -119,8 +119,8 @@
{ {
"pdf": "011_moyen_compte_rendu_CRH_23080179.pdf", "pdf": "011_moyen_compte_rendu_CRH_23080179.pdf",
"success": true, "success": true,
"time_s": 1.0042967796325684, "time_s": 0.9325697422027588,
"pii_count": 46, "pii_count": 20,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/011_moyen_compte_rendu_CRH_23080179.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/011_moyen_compte_rendu_CRH_23080179.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/011_moyen_compte_rendu_CRH_23080179.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/011_moyen_compte_rendu_CRH_23080179.audit.jsonl",
@@ -131,8 +131,8 @@
{ {
"pdf": "012_moyen_compte_rendu_CRH_692_23200418.pdf", "pdf": "012_moyen_compte_rendu_CRH_692_23200418.pdf",
"success": true, "success": true,
"time_s": 0.8403730392456055, "time_s": 0.6736557483673096,
"pii_count": 103, "pii_count": 32,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/012_moyen_compte_rendu_CRH_692_23200418.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/012_moyen_compte_rendu_CRH_692_23200418.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/012_moyen_compte_rendu_CRH_692_23200418.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/012_moyen_compte_rendu_CRH_692_23200418.audit.jsonl",
@@ -143,8 +143,8 @@
{ {
"pdf": "013_moyen_compte_rendu_363_23085243_CRO.pdf", "pdf": "013_moyen_compte_rendu_363_23085243_CRO.pdf",
"success": true, "success": true,
"time_s": 0.94016432762146, "time_s": 0.6802682876586914,
"pii_count": 160, "pii_count": 34,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/013_moyen_compte_rendu_363_23085243_CRO.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/013_moyen_compte_rendu_363_23085243_CRO.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/013_moyen_compte_rendu_363_23085243_CRO.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/013_moyen_compte_rendu_363_23085243_CRO.audit.jsonl",
@@ -155,7 +155,7 @@
{ {
"pdf": "014_moyen_compte_rendu_CRO_23167029.redacted_raster.pdf", "pdf": "014_moyen_compte_rendu_CRO_23167029.redacted_raster.pdf",
"success": true, "success": true,
"time_s": 0.4384956359863281, "time_s": 0.4354434013366699,
"pii_count": 0, "pii_count": 0,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/014_moyen_compte_rendu_CRO_23167029.redacted_raster.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/014_moyen_compte_rendu_CRO_23167029.redacted_raster.pseudonymise.txt",
@@ -167,8 +167,8 @@
{ {
"pdf": "015_moyen_unknown_CONSULTATION_ANESTHESISTE_23139653.pdf", "pdf": "015_moyen_unknown_CONSULTATION_ANESTHESISTE_23139653.pdf",
"success": true, "success": true,
"time_s": 0.9846677780151367, "time_s": 0.9319710731506348,
"pii_count": 25, "pii_count": 7,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/015_moyen_unknown_CONSULTATION_ANESTHESISTE_23139653.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/015_moyen_unknown_CONSULTATION_ANESTHESISTE_23139653.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/015_moyen_unknown_CONSULTATION_ANESTHESISTE_23139653.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/015_moyen_unknown_CONSULTATION_ANESTHESISTE_23139653.audit.jsonl",
@@ -179,8 +179,8 @@
{ {
"pdf": "016_moyen_compte_rendu_CRH_23149905.pdf", "pdf": "016_moyen_compte_rendu_CRH_23149905.pdf",
"success": true, "success": true,
"time_s": 1.4508278369903564, "time_s": 1.150942325592041,
"pii_count": 242, "pii_count": 117,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/016_moyen_compte_rendu_CRH_23149905.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/016_moyen_compte_rendu_CRH_23149905.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/016_moyen_compte_rendu_CRH_23149905.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/016_moyen_compte_rendu_CRH_23149905.audit.jsonl",
@@ -191,7 +191,7 @@
{ {
"pdf": "017_moyen_compte_rendu_CRO_23222062.redacted_raster.pdf", "pdf": "017_moyen_compte_rendu_CRO_23222062.redacted_raster.pdf",
"success": true, "success": true,
"time_s": 0.4350569248199463, "time_s": 0.43438720703125,
"pii_count": 0, "pii_count": 0,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/017_moyen_compte_rendu_CRO_23222062.redacted_raster.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/017_moyen_compte_rendu_CRO_23222062.redacted_raster.pseudonymise.txt",
@@ -203,8 +203,8 @@
{ {
"pdf": "018_moyen_compte_rendu_CRH_23042753.pdf", "pdf": "018_moyen_compte_rendu_CRH_23042753.pdf",
"success": true, "success": true,
"time_s": 1.9062294960021973, "time_s": 1.5716781616210938,
"pii_count": 233, "pii_count": 123,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/018_moyen_compte_rendu_CRH_23042753.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/018_moyen_compte_rendu_CRH_23042753.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/018_moyen_compte_rendu_CRH_23042753.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/018_moyen_compte_rendu_CRH_23042753.audit.jsonl",
@@ -215,8 +215,8 @@
{ {
"pdf": "019_moyen_compte_rendu_CRO_332_23049003.pdf", "pdf": "019_moyen_compte_rendu_CRO_332_23049003.pdf",
"success": true, "success": true,
"time_s": 1.020752191543579, "time_s": 0.7931430339813232,
"pii_count": 161, "pii_count": 71,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/019_moyen_compte_rendu_CRO_332_23049003.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/019_moyen_compte_rendu_CRO_332_23049003.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/019_moyen_compte_rendu_CRO_332_23049003.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/019_moyen_compte_rendu_CRO_332_23049003.audit.jsonl",
@@ -227,7 +227,7 @@
{ {
"pdf": "020_moyen_compte_rendu_CRO_23084754.redacted_raster.pdf", "pdf": "020_moyen_compte_rendu_CRO_23084754.redacted_raster.pdf",
"success": true, "success": true,
"time_s": 0.4804375171661377, "time_s": 0.43088579177856445,
"pii_count": 0, "pii_count": 0,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/020_moyen_compte_rendu_CRO_23084754.redacted_raster.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/020_moyen_compte_rendu_CRO_23084754.redacted_raster.pseudonymise.txt",
@@ -239,7 +239,7 @@
{ {
"pdf": "021_moyen_compte_rendu_CRO_23201117.redacted_raster.pdf", "pdf": "021_moyen_compte_rendu_CRO_23201117.redacted_raster.pdf",
"success": true, "success": true,
"time_s": 0.31412649154663086, "time_s": 0.3120863437652588,
"pii_count": 0, "pii_count": 0,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/021_moyen_compte_rendu_CRO_23201117.redacted_raster.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/021_moyen_compte_rendu_CRO_23201117.redacted_raster.pseudonymise.txt",
@@ -251,8 +251,8 @@
{ {
"pdf": "022_moyen_compte_rendu_cro2_516_23187028.pdf", "pdf": "022_moyen_compte_rendu_cro2_516_23187028.pdf",
"success": true, "success": true,
"time_s": 0.37198877334594727, "time_s": 0.35700511932373047,
"pii_count": 29, "pii_count": 4,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/022_moyen_compte_rendu_cro2_516_23187028.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/022_moyen_compte_rendu_cro2_516_23187028.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/022_moyen_compte_rendu_cro2_516_23187028.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/022_moyen_compte_rendu_cro2_516_23187028.audit.jsonl",
@@ -263,8 +263,8 @@
{ {
"pdf": "023_complexe_compte_rendu_CRH_23102610.pdf", "pdf": "023_complexe_compte_rendu_CRH_23102610.pdf",
"success": true, "success": true,
"time_s": 4.054161310195923, "time_s": 2.7280702590942383,
"pii_count": 617, "pii_count": 385,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/023_complexe_compte_rendu_CRH_23102610.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/023_complexe_compte_rendu_CRH_23102610.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/023_complexe_compte_rendu_CRH_23102610.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/023_complexe_compte_rendu_CRH_23102610.audit.jsonl",
@@ -275,8 +275,8 @@
{ {
"pdf": "024_complexe_trackare_trackare-17001141-23066188_17001141_23066188.pdf", "pdf": "024_complexe_trackare_trackare-17001141-23066188_17001141_23066188.pdf",
"success": true, "success": true,
"time_s": 8.550535917282104, "time_s": 5.714028835296631,
"pii_count": 804, "pii_count": 117,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/024_complexe_trackare_trackare-17001141-23066188_17001141_23066188.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/024_complexe_trackare_trackare-17001141-23066188_17001141_23066188.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/024_complexe_trackare_trackare-17001141-23066188_17001141_23066188.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/024_complexe_trackare_trackare-17001141-23066188_17001141_23066188.audit.jsonl",
@@ -287,8 +287,8 @@
{ {
"pdf": "025_complexe_trackare_trackare-02016820-23095226_02016820_23095226.pdf", "pdf": "025_complexe_trackare_trackare-02016820-23095226_02016820_23095226.pdf",
"success": true, "success": true,
"time_s": 17.83988666534424, "time_s": 9.729689836502075,
"pii_count": 1622, "pii_count": 270,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/025_complexe_trackare_trackare-02016820-23095226_02016820_23095226.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/025_complexe_trackare_trackare-02016820-23095226_02016820_23095226.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/025_complexe_trackare_trackare-02016820-23095226_02016820_23095226.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/025_complexe_trackare_trackare-02016820-23095226_02016820_23095226.audit.jsonl",
@@ -299,8 +299,8 @@
{ {
"pdf": "026_complexe_trackare_trackare-15000536-23074384_15000536_23074384.pdf", "pdf": "026_complexe_trackare_trackare-15000536-23074384_15000536_23074384.pdf",
"success": true, "success": true,
"time_s": 12.040966749191284, "time_s": 7.467007637023926,
"pii_count": 1056, "pii_count": 142,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/026_complexe_trackare_trackare-15000536-23074384_15000536_23074384.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/026_complexe_trackare_trackare-15000536-23074384_15000536_23074384.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/026_complexe_trackare_trackare-15000536-23074384_15000536_23074384.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/026_complexe_trackare_trackare-15000536-23074384_15000536_23074384.audit.jsonl",
@@ -311,8 +311,8 @@
{ {
"pdf": "027_complexe_trackare_trackare-10027557-23183041_10027557_23183041.pdf", "pdf": "027_complexe_trackare_trackare-10027557-23183041_10027557_23183041.pdf",
"success": true, "success": true,
"time_s": 8.782238721847534, "time_s": 6.15097975730896,
"pii_count": 859, "pii_count": 121,
"files": { "files": {
"text": "tests/ground_truth/pdfs/baseline_anonymized/027_complexe_trackare_trackare-10027557-23183041_10027557_23183041.pseudonymise.txt", "text": "tests/ground_truth/pdfs/baseline_anonymized/027_complexe_trackare_trackare-10027557-23183041_10027557_23183041.pseudonymise.txt",
"audit": "tests/ground_truth/pdfs/baseline_anonymized/027_complexe_trackare_trackare-10027557-23183041_10027557_23183041.audit.jsonl", "audit": "tests/ground_truth/pdfs/baseline_anonymized/027_complexe_trackare_trackare-10027557-23183041_10027557_23183041.audit.jsonl",

View File

@@ -2,11 +2,11 @@
"evaluation_date": "2026-03-02", "evaluation_date": "2026-03-02",
"total_documents": 25, "total_documents": 25,
"global_metrics": { "global_metrics": {
"precision": 0.1897, "precision": 0.8827,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.3189, "f1_score": 0.9377,
"true_positives": 1159, "true_positives": 1159,
"false_positives": 4951, "false_positives": 154,
"false_negatives": 0 "false_negatives": 0
}, },
"by_type": { "by_type": {
@@ -42,38 +42,6 @@
"false_positives": 0, "false_positives": 0,
"false_negatives": 0 "false_negatives": 0
}, },
"NOM_EXTRACTED": {
"precision": 0.0,
"recall": 0.0,
"f1_score": 0.0,
"true_positives": 0,
"false_positives": 3846,
"false_negatives": 0
},
"NOM_GLOBAL": {
"precision": 0.0,
"recall": 0.0,
"f1_score": 0.0,
"true_positives": 0,
"false_positives": 670,
"false_negatives": 0
},
"ETAB_GLOBAL": {
"precision": 0.0,
"recall": 0.0,
"f1_score": 0.0,
"true_positives": 0,
"false_positives": 36,
"false_negatives": 0
},
"TEL_GLOBAL": {
"precision": 0.0,
"recall": 0.0,
"f1_score": 0.0,
"true_positives": 0,
"false_positives": 77,
"false_negatives": 0
},
"ADRESSE": { "ADRESSE": {
"precision": 0.878, "precision": 0.878,
"recall": 1.0, "recall": 1.0,
@@ -98,30 +66,6 @@
"false_positives": 0, "false_positives": 0,
"false_negatives": 0 "false_negatives": 0
}, },
"ADRESSE_GLOBAL": {
"precision": 0.0,
"recall": 0.0,
"f1_score": 0.0,
"true_positives": 0,
"false_positives": 55,
"false_negatives": 0
},
"CODE_POSTAL_GLOBAL": {
"precision": 0.0,
"recall": 0.0,
"f1_score": 0.0,
"true_positives": 0,
"false_positives": 39,
"false_negatives": 0
},
"DATE_NAISSANCE_GLOBAL": {
"precision": 0.0,
"recall": 0.0,
"f1_score": 0.0,
"true_positives": 0,
"false_positives": 20,
"false_negatives": 0
},
"EMAIL": { "EMAIL": {
"precision": 1.0, "precision": 1.0,
"recall": 1.0, "recall": 1.0,
@@ -146,30 +90,6 @@
"false_positives": 106, "false_positives": 106,
"false_negatives": 0 "false_negatives": 0
}, },
"EMAIL_GLOBAL": {
"precision": 0.0,
"recall": 0.0,
"f1_score": 0.0,
"true_positives": 0,
"false_positives": 28,
"false_negatives": 0
},
"RPPS_GLOBAL": {
"precision": 0.0,
"recall": 0.0,
"f1_score": 0.0,
"true_positives": 0,
"false_positives": 7,
"false_negatives": 0
},
"EPISODE_GLOBAL": {
"precision": 0.0,
"recall": 0.0,
"f1_score": 0.0,
"true_positives": 0,
"false_positives": 9,
"false_negatives": 0
},
"VILLE": { "VILLE": {
"precision": 0.2, "precision": 0.2,
"recall": 1.0, "recall": 1.0,
@@ -178,14 +98,6 @@
"false_positives": 20, "false_positives": 20,
"false_negatives": 0 "false_negatives": 0
}, },
"VILLE_GLOBAL": {
"precision": 0.0,
"recall": 0.0,
"f1_score": 0.0,
"true_positives": 0,
"false_positives": 10,
"false_negatives": 0
},
"AGE": { "AGE": {
"precision": 1.0, "precision": 1.0,
"recall": 1.0, "recall": 1.0,
@@ -214,29 +126,29 @@
"per_document": [ "per_document": [
{ {
"pdf": "001_simple_unknown_BACTERIO_23018396", "pdf": "001_simple_unknown_BACTERIO_23018396",
"precision": 0.2326, "precision": 1.0,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.3774, "f1_score": 1.0,
"true_positives": 10, "true_positives": 10,
"false_positives": 33, "false_positives": 0,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
"pdf": "002_simple_unknown_bacterio_476_23159413", "pdf": "002_simple_unknown_bacterio_476_23159413",
"precision": 0.234, "precision": 1.0,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.3793, "f1_score": 1.0,
"true_positives": 11, "true_positives": 11,
"false_positives": 36, "false_positives": 0,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
"pdf": "003_simple_compte_rendu_CRO_23155084", "pdf": "003_simple_compte_rendu_CRO_23155084",
"precision": 0.16, "precision": 1.0,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.2759, "f1_score": 1.0,
"true_positives": 4, "true_positives": 4,
"false_positives": 21, "false_positives": 0,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
@@ -250,65 +162,65 @@
}, },
{ {
"pdf": "005_simple_compte_rendu_CRH_23155836", "pdf": "005_simple_compte_rendu_CRH_23155836",
"precision": 0.4429, "precision": 1.0,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.6139, "f1_score": 1.0,
"true_positives": 62, "true_positives": 62,
"false_positives": 78, "false_positives": 0,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
"pdf": "008_simple_trackare_trackare-14004105-23202435_14004105_23202435", "pdf": "008_simple_trackare_trackare-14004105-23202435_14004105_23202435",
"precision": 0.1899, "precision": 0.5769,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.3191, "f1_score": 0.7317,
"true_positives": 15, "true_positives": 15,
"false_positives": 64, "false_positives": 11,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
"pdf": "009_simple_compte_rendu_CRO_23051225", "pdf": "009_simple_compte_rendu_CRO_23051225",
"precision": 0.25, "precision": 1.0,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.4, "f1_score": 1.0,
"true_positives": 8, "true_positives": 8,
"false_positives": 24, "false_positives": 0,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
"pdf": "010_simple_anapath_ANAPATH_23217289", "pdf": "010_simple_anapath_ANAPATH_23217289",
"precision": 0.2549, "precision": 1.0,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.4062, "f1_score": 1.0,
"true_positives": 13, "true_positives": 13,
"false_positives": 38, "false_positives": 0,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
"pdf": "011_moyen_compte_rendu_CRH_23080179", "pdf": "011_moyen_compte_rendu_CRH_23080179",
"precision": 0.3158, "precision": 1.0,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.48, "f1_score": 1.0,
"true_positives": 12, "true_positives": 12,
"false_positives": 26, "false_positives": 0,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
"pdf": "012_moyen_compte_rendu_CRH_692_23200418", "pdf": "012_moyen_compte_rendu_CRH_692_23200418",
"precision": 0.297, "precision": 1.0,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.458, "f1_score": 1.0,
"true_positives": 30, "true_positives": 30,
"false_positives": 71, "false_positives": 0,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
"pdf": "013_moyen_compte_rendu_363_23085243_CRO", "pdf": "013_moyen_compte_rendu_363_23085243_CRO",
"precision": 0.2025, "precision": 1.0,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.3368, "f1_score": 1.0,
"true_positives": 32, "true_positives": 32,
"false_positives": 126, "false_positives": 0,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
@@ -322,20 +234,20 @@
}, },
{ {
"pdf": "015_moyen_unknown_CONSULTATION_ANESTHESISTE_23139653", "pdf": "015_moyen_unknown_CONSULTATION_ANESTHESISTE_23139653",
"precision": 0.28, "precision": 1.0,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.4375, "f1_score": 1.0,
"true_positives": 7, "true_positives": 7,
"false_positives": 18, "false_positives": 0,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
"pdf": "016_moyen_compte_rendu_CRH_23149905", "pdf": "016_moyen_compte_rendu_CRH_23149905",
"precision": 0.477, "precision": 1.0,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.6459, "f1_score": 1.0,
"true_positives": 114, "true_positives": 114,
"false_positives": 125, "false_positives": 0,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
@@ -349,20 +261,20 @@
}, },
{ {
"pdf": "018_moyen_compte_rendu_CRH_23042753", "pdf": "018_moyen_compte_rendu_CRH_23042753",
"precision": 0.5279, "precision": 1.0,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.691, "f1_score": 1.0,
"true_positives": 123, "true_positives": 123,
"false_positives": 110, "false_positives": 0,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
"pdf": "019_moyen_compte_rendu_CRO_332_23049003", "pdf": "019_moyen_compte_rendu_CRO_332_23049003",
"precision": 0.3793, "precision": 1.0,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.55, "f1_score": 1.0,
"true_positives": 55, "true_positives": 55,
"false_positives": 90, "false_positives": 0,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
@@ -385,56 +297,56 @@
}, },
{ {
"pdf": "022_moyen_compte_rendu_cro2_516_23187028", "pdf": "022_moyen_compte_rendu_cro2_516_23187028",
"precision": 0.1379, "precision": 1.0,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.2424, "f1_score": 1.0,
"true_positives": 4, "true_positives": 4,
"false_positives": 25, "false_positives": 0,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
"pdf": "023_complexe_compte_rendu_CRH_23102610", "pdf": "023_complexe_compte_rendu_CRH_23102610",
"precision": 0.6203, "precision": 1.0,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.7657, "f1_score": 1.0,
"true_positives": 379, "true_positives": 379,
"false_positives": 232, "false_positives": 0,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
"pdf": "024_complexe_trackare_trackare-17001141-23066188_17001141_23066188", "pdf": "024_complexe_trackare_trackare-17001141-23066188_17001141_23066188",
"precision": 0.0689, "precision": 0.6463,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.129, "f1_score": 0.7852,
"true_positives": 53, "true_positives": 53,
"false_positives": 716, "false_positives": 29,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
"pdf": "025_complexe_trackare_trackare-02016820-23095226_02016820_23095226", "pdf": "025_complexe_trackare_trackare-02016820-23095226_02016820_23095226",
"precision": 0.0643, "precision": 0.6857,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.1209, "f1_score": 0.8136,
"true_positives": 96, "true_positives": 96,
"false_positives": 1396, "false_positives": 44,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
"pdf": "026_complexe_trackare_trackare-15000536-23074384_15000536_23074384", "pdf": "026_complexe_trackare_trackare-15000536-23074384_15000536_23074384",
"precision": 0.0766, "precision": 0.6695,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.1422, "f1_score": 0.802,
"true_positives": 79, "true_positives": 79,
"false_positives": 953, "false_positives": 39,
"false_negatives": 0 "false_negatives": 0
}, },
{ {
"pdf": "027_complexe_trackare_trackare-10027557-23183041_10027557_23183041", "pdf": "027_complexe_trackare_trackare-10027557-23183041_10027557_23183041",
"precision": 0.0633, "precision": 0.6265,
"recall": 1.0, "recall": 1.0,
"f1_score": 0.1191, "f1_score": 0.7704,
"true_positives": 52, "true_positives": 52,
"false_positives": 769, "false_positives": 31,
"false_negatives": 0 "false_negatives": 0
} }
] ]