File size: 17,066 Bytes
c8b6043 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 | """
UnifiedTurkishEntityEngine — `ictihat_core` × `kvkk-pii` × `TürkçeOCR` Birleşik Varlık Tanıma,
Hiyerarşik Span Kapsama (Hierarchical Span Subsumption) ve KVKK Karartma Motoru.
"""
from __future__ import annotations
import re
import sys
from dataclasses import dataclass, field
from pathlib import Path
from typing import Any, Dict, List, Optional, Set, Tuple
from turkceocr.postprocess.legal_recognizers import (
DetectedEntitySpan,
LegalAndCommercialRecognizerSuite,
)
try:
from kvkk_pii.engine import PIIAnalyzer as _KvkkPIIAnalyzer # type: ignore
except Exception:
_KvkkPIIAnalyzer = None
# Hiyerarşik kapsayıcı varlık tipleri (içlerindeki kısa tarih/plaka/sayı spanları kapsayıcıyı sildirtmez!)
CONTAINER_ENTITY_TYPES: Set[str] = {
"ADDRESS",
"HEALTH_DATA",
"KVKK_SPECIAL",
"ORG_COMPANY",
"TAX_OFFICE",
"TR_COURT_AUTHORITY",
"TR_LAW_CITATION",
}
# KVKK kapsamında karartılması (redaction) zorunlu olan hassas PII + SUSPECT tipleri
DEFAULT_REDACTABLE_ENTITY_TYPES: Set[str] = {
"TR_TCKN",
"TR_TCKN_SUSPECT",
"TR_YABANCI_KIMLIK",
"TR_VKN",
"TR_VKN_SUSPECT",
"TR_MERSIS",
"TR_IBAN",
"TR_IBAN_SUSPECT",
"TR_PHONE",
"EMAIL",
"TR_EMAIL",
"TR_KEP",
"TR_PLATE",
"CREDIT_CARD",
"TR_CREDIT_CARD",
"TR_PASSPORT",
"IP_ADDRESS",
"TR_IP",
"TR_IPV6",
"TR_MAC",
"SECRET",
"TR_SECRET",
"PERSON",
"ADDRESS",
"HEALTH_DATA",
"KVKK_SPECIAL",
"TR_SGK_SICIL",
"TR_TAPU_ADA_PARSEL",
}
# Bağlamsal hibrit tanıyıcı kalıpları (NER modeli yüklü olmadığında veya NER'i tamamlayıcı olarak)
_RX_TAX_OFFICE = re.compile(
r"\b([A-ZÇĞİÖŞÜ][a-zçğıöşüA-ZÇĞİÖŞÜ]{1,20}(?:[^\S\r\n]+[A-ZÇĞİÖŞÜa-zçğıöşü]{1,20}){0,3}[^\S\r\n]+Vergi[^\S\r\n]+Dairesi(?:[^\S\r\n]+Müdürlüğü)?)\b"
)
_RX_ORG_COMPANY = re.compile(
r"\b([A-ZÇĞİÖŞÜ0-9][A-Za-zÇĞİÖŞÜçğıöşü0-9\.\-&]{1,25}(?:[^\S\r\n]+[A-Za-zÇĞİÖŞÜçğıöşü0-9\.\-&]{1,25}){0,7}[^\S\r\n]+"
r"(?:T\.?[^\S\r\n]*A\.?[^\S\r\n]*Ş\.?|T\.?[^\S\r\n]*A\.?[^\S\r\n]*O\.?|A\.?[^\S\r\n]*Ş\.?|A\.?[^\S\r\n]*O\.?|LTD\.?[^\S\r\n]*ŞTİ\.?|LİMİTED[^\S\r\n]+ŞİRKETİ|ANONİM[^\S\r\n]+ŞİRKETİ|SAN\.?[^\S\r\n]*(?:VE[^\S\r\n]*)?TİC\.?[^\S\r\n]*(?:T\.?[^\S\r\n]*A\.?[^\S\r\n]*Ş\.?|A\.?[^\S\r\n]*Ş\.?|LTD\.?[^\S\r\n]*ŞTİ\.?)|HOLDİNG[^\S\r\n]+A\.?[^\S\r\n]*Ş\.?))"
)
_RX_LABELED_PERSON = re.compile(
r"(?:(?:Ad[ıi][^\S\r\n]*Soyad[ıi]|Müşteri(?:[^\S\r\n]*Ad[ıi])?|Hasta(?:[^\S\r\n]*Ad[ıi])?|Alıcı|Borçlu|Alacaklı|Davacı|Davalı|Başvurucu|Sanık|Müşteki|Vekili|Av\.|Avukat|Sayın|Sn\.)"
r"(?:[^\S\r\n]+(?:Gerçek[^\S\r\n]+Kişi|Ad[ıi][^\S\r\n]*Soyad[ıi]))?[^\S\r\n]*[:\-]?[^\S\r\n]*)"
r"([A-ZÇĞİÖŞÜ][a-zçğıöşüA-ZÇĞİÖŞÜ]{1,20}(?:[^\S\r\n]+[A-ZÇĞİÖŞÜ][a-zçğıöşüA-ZÇĞİÖŞÜ]{1,22}){1,3})"
)
_RX_LABELED_ADDRESS = re.compile(
r"(?:(?:İletişim[^\S\r\n]*&[^\S\r\n]*Adres|Adres[i]?|İkametgah|Teslimat[^\S\r\n]*Adresi|Sevk[^\S\r\n]*Adresi)[^\S\r\n]*[:\-][^\S\r\n]*)"
r"([^\n\r\|]{10,120}?)(?=\s*\||(?<!Mah)(?<!Cad)(?<!Sok)(?<!Sk)(?<!Blv)(?<!Apt)(?<!No)(?<!Kat)\.\s+[A-ZÇĞİÖŞÜ]|[\n\r]|$)"
)
_RX_UNLABELED_ADDRESS = re.compile(
r"\b([A-ZÇĞİÖŞÜ][a-zçğıöşüA-ZÇĞİÖŞÜ]{1,22}(?:[^\S\r\n]+[A-ZÇĞİÖŞÜa-zçğıöşü]{1,22}){0,2}[^\S\r\n]+(?:Mahallesi|Mah\.|Bulvarı|Blv\.|Caddesi|Cad\.|Sokağı|Sok\.|Sk\.)[^\S\r\n]+[^\n\r\|]{0,95}?(?:(?:\d{1,3}[^\S\r\n]*[/\/][^\S\r\n]*[A-Za-zÇĞİÖŞÜçğıöşü]+)|(?:Kat[^\S\r\n]*[:\.]?[^\S\r\n]*\d+)|(?:Daire[^\S\r\n]*[:\.]?[^\S\r\n]*\d+)|(?:No[^\S\r\n]*[:\.]?[^\S\r\n]*\d+[A-Za-z\/0-9\s]*[A-Za-zÇĞİÖŞÜçğıöşü]+\s*/\s*[A-Za-zÇĞİÖŞÜçğıöşü]+)|(?:No[^\S\r\n]*[:\.]?[^\S\r\n]*\d+[A-Za-z\/0-9]*)))"
)
_RX_HEALTH_DATA = re.compile(
r"(?:(?:Sağlık[^\S\r\n]*/[^\S\r\n]*Epikriz|Tanı|Teşhis|ICD(?:\-10)?(?:\s*Kodu)?|Reçete|İlaç|Kan\s*Grubu|Epikriz|Poliklinik\s*Notu)\s*[:\-]\s*)"
r"([^\n\r\|]{5,110}?)(?=\s*\||[\n\r]|$)"
r"|(\b[A-Z]\d{2}\.\d\s*[-–]?\s*[A-Za-zÇĞİÖŞÜçğıöşü0-9\s]{4,65})"
r"|(\b(?:0|A|B|AB)\s*Rh\s*[\+\-]\s*\((?:Pozitif|Negatif)\))"
)
_RX_KVKK_SPECIAL = re.compile(
r"(?:(?:KVKK[^\S\r\n]*Md\.?\s*6[^\S\r\n]*Özel[^\S\r\n]*Veri\s*[:\-]\s*)?"
r"(?:Dini(?:\s*İnancı)?|Mezhebi|Etnik\s*Kökeni?|Siyasi\s*Düşüncesi|Sendika(?:\s*Üyeliği)?|Dernek(?:\s*Üyeliği)?|Sabıka\s*Kaydı|Adli\s*Sicil|Ceza\s*Mahk[uû]miyeti|Biyometrik\s*Veri)\s*[:\-]\s*)"
r"([^\n\r\|]{4,100}?)(?=\s*\||[\n\r]|$)"
r"|(\b[A-ZÇĞİÖŞÜa-zçğıöşü\s]{3,35}Sendikası(?:\s+üyeliği)?(?:\s+ve\s+adli\s+sicil\s+sabıka\s+kaydı)?)"
)
@dataclass
class UnifiedAnalysisReport:
"""Birleşik Varlık Tanıma ve KVKK Analiz Raporu."""
text: str
entities: List[DetectedEntitySpan] = field(default_factory=list)
redacted_text: str = ""
entity_counts: Dict[str, int] = field(default_factory=dict)
def to_dict(self) -> Dict[str, Any]:
return {
"text_length": len(self.text),
"entity_count": len(self.entities),
"entity_counts": self.entity_counts,
"entities": [e.to_dict() for e in self.entities],
"redacted_text": self.redacted_text,
}
class UnifiedTurkishEntityEngine:
"""
`ictihat_core` (Hukuki Tier-1) + `kvkk-pii` (Kişisel Veri Tier-1 + BERTurkNER v3) +
`TürkçeOCR` (Ticari Tier-1 + İki Katmanlı OCR-Toleranslı Fail-Closed Dedektör) birleşik motoru.
"""
def __init__(
self,
use_ner: bool = False,
ner_model_path: Optional[str] = None,
score_threshold: float = 0.50,
) -> None:
self.use_ner = use_ner
self.ner_model_path = ner_model_path
self.score_threshold = score_threshold
self.legal_suite = LegalAndCommercialRecognizerSuite()
self.kvkk_analyzer = None
if _KvkkPIIAnalyzer is not None:
try:
self.kvkk_analyzer = _KvkkPIIAnalyzer(ner_model_path=ner_model_path)
except Exception:
self.kvkk_analyzer = _KvkkPIIAnalyzer()
def _extract_contextual_entities(self, text: str) -> List[DetectedEntitySpan]:
spans: List[DetectedEntitySpan] = []
for m in _RX_TAX_OFFICE.finditer(text):
s, e = m.start(1), m.end(1)
spans.append(
DetectedEntitySpan(
entity_type="TAX_OFFICE",
start=s,
end=e,
score=0.95,
text=text[s:e],
source="contextual_grammar",
tier="VERIFIED",
)
)
for m in _RX_ORG_COMPANY.finditer(text):
s, e = m.start(1), m.end(1)
cand_text = text[s:e].strip()
# Mahkeme/Daire başlığı değilse şirket olarak ekle
if "Mahkemesi" not in cand_text and "Dairesi" not in cand_text:
spans.append(
DetectedEntitySpan(
entity_type="ORG_COMPANY",
start=s,
end=e,
score=0.93,
text=text[s:e],
source="contextual_grammar",
tier="VERIFIED",
)
)
for m in _RX_LABELED_PERSON.finditer(text):
s, e = m.start(1), m.end(1)
cand_text = text[s:e].strip()
if not any(
bad in cand_text
for bad in ("Vergi Dairesi", "Mahkemesi", "Dairesi", "Kanunu", "Müdürlüğü", "Şirketi")
):
spans.append(
DetectedEntitySpan(
entity_type="PERSON",
start=s,
end=e,
score=0.92,
text=text[s:e],
source="contextual_grammar",
tier="VERIFIED_PII",
)
)
for m in _RX_LABELED_ADDRESS.finditer(text):
s, e = m.start(1), m.end(1)
spans.append(
DetectedEntitySpan(
entity_type="ADDRESS",
start=s,
end=e,
score=0.92,
text=text[s:e],
source="contextual_grammar",
tier="VERIFIED_PII",
)
)
for m in _RX_UNLABELED_ADDRESS.finditer(text):
s, e = m.start(1), m.end(1)
spans.append(
DetectedEntitySpan(
entity_type="ADDRESS",
start=s,
end=e,
score=0.90,
text=text[s:e],
source="contextual_grammar",
tier="VERIFIED_PII",
)
)
for m in _RX_HEALTH_DATA.finditer(text):
s, e = m.start(), m.end()
for g_idx in (1, 2, 3):
if m.group(g_idx) is not None:
s, e = m.start(g_idx), m.end(g_idx)
break
spans.append(
DetectedEntitySpan(
entity_type="HEALTH_DATA",
start=s,
end=e,
score=0.93,
text=text[s:e],
source="contextual_grammar",
tier="VERIFIED_PII",
)
)
for m in _RX_KVKK_SPECIAL.finditer(text):
s, e = m.start(), m.end()
for g_idx in (1, 2):
if m.group(g_idx) is not None:
s, e = m.start(g_idx), m.end(g_idx)
break
spans.append(
DetectedEntitySpan(
entity_type="KVKK_SPECIAL",
start=s,
end=e,
score=0.93,
text=text[s:e],
source="contextual_grammar",
tier="VERIFIED_PII",
)
)
return spans
@staticmethod
def deduplicate_hierarchical(entities: List[DetectedEntitySpan]) -> List[DetectedEntitySpan]:
"""
Hiyerarşik Span Kapsama (Hierarchical Span Subsumption):
1. `TR_COURT_AUTHORITY` (`score=1.0`) kendisiyle örtüşen yanlış pozitif `ORG_COMPANY`,
`ADDRESS` veya `PERSON` spanlarını ezer.
2. Kapsayıcı bir varlık (`ADDRESS`, `HEALTH_DATA`, `KVKK_SPECIAL`, `ORG_COMPANY`,
`TAX_OFFICE`, `TR_COURT_AUTHORITY`, `TR_LAW_CITATION`) kendisinden daha kısa bir
alt-span (`TR_DATE`, `TR_PLATE`, `TR_ESAS_KARAR_NO` vb.) ile örtüştüğünde kapsayıcı
varlık ASLA silinmez; kısa alt-span `container.sub_entities` listesine eklenir.
"""
if not entities:
return []
sorted_ents = sorted(
entities,
key=lambda e: (e.start, -(e.end - e.start), -e.score),
)
kept: List[DetectedEntitySpan] = []
for candidate in sorted_ents:
if not kept:
kept.append(candidate)
continue
prev = kept[-1]
if candidate.start < prev.end:
prev_len = prev.end - prev.start
cand_len = candidate.end - candidate.start
# Özel kural: TR_COURT_AUTHORITY her zaman ORG_COMPANY / ADDRESS / PERSON yanlış pozitifini ezer
if prev.entity_type == "TR_COURT_AUTHORITY" and candidate.entity_type in {
"ORG_COMPANY",
"ADDRESS",
"PERSON",
}:
continue
if candidate.entity_type == "TR_COURT_AUTHORITY" and prev.entity_type in {
"ORG_COMPANY",
"ADDRESS",
"PERSON",
}:
kept[-1] = candidate
continue
# Özel kural: TR_MERSIS (16 hane) içindeki 10 haneli TR_VKN veya TR_PHONE'u alt-varlık olarak yutar
if prev.entity_type == "TR_MERSIS" and candidate.entity_type in {"TR_VKN", "TR_PHONE"}:
prev.sub_entities.append(candidate)
continue
if candidate.entity_type == "TR_MERSIS" and prev.entity_type in {"TR_VKN", "TR_PHONE"}:
candidate.sub_entities.append(prev)
kept[-1] = candidate
continue
prev_is_container = prev.entity_type in CONTAINER_ENTITY_TYPES
cand_is_container = candidate.entity_type in CONTAINER_ENTITY_TYPES
if (
prev_is_container
and not cand_is_container
and candidate.start >= prev.start
and candidate.end <= prev.end
and prev_len > cand_len
):
prev.sub_entities.append(candidate)
continue
if (
cand_is_container
and not prev_is_container
and prev.start >= candidate.start
and prev.end <= candidate.end
and cand_len > prev_len
):
candidate.sub_entities.append(prev)
kept[-1] = candidate
continue
if candidate.score > prev.score or (
candidate.score == prev.score and cand_len > prev_len
):
kept[-1] = candidate
else:
kept.append(candidate)
return kept
def analyze(
self,
text: str,
redact_types: Optional[Set[str]] = None,
) -> UnifiedAnalysisReport:
if not text:
return UnifiedAnalysisReport(text="", entities=[], redacted_text="")
raw_entities: List[DetectedEntitySpan] = []
# 1. kvkk-pii çekirdek tanıyıcıları
if self.kvkk_analyzer is not None:
for e in self.kvkk_analyzer.analyze(text):
if isinstance(e, dict):
raw_entities.append(
DetectedEntitySpan(
entity_type=str(e["entity_type"]),
start=int(e["start"]),
end=int(e["end"]),
score=float(e["score"]),
text=str(e["text"]),
source=str(e.get("source", "kvkk_pii")),
tier="VERIFIED_PII",
redaction_token=f"[{e['entity_type']}]",
)
)
else:
raw_entities.append(
DetectedEntitySpan(
entity_type=e.entity_type,
start=e.start,
end=e.end,
score=e.score,
text=e.text,
source=e.source,
tier="VERIFIED_PII",
redaction_token=f"[{e.entity_type}]",
)
)
# 2. Hukuki (`ictihat_core`), Ticari ve Fail-Closed OCR-Toleranslı (`VERIFIED_PII` + `SUSPECT_PII`) tanıyıcılar
raw_entities.extend(self.legal_suite.analyze(text))
# 3. Bağlamsal hibrit tanıyıcılar (`PERSON`, `ADDRESS`, `HEALTH_DATA`, `KVKK_SPECIAL`, `ORG_COMPANY`, `TAX_OFFICE`)
raw_entities.extend(self._extract_contextual_entities(text))
# Eşik filtreleme ve Hiyerarşik Span Kapsama
filtered = [e for e in raw_entities if e.score >= self.score_threshold]
deduped = self.deduplicate_hierarchical(filtered)
counts: Dict[str, int] = {}
for ent in deduped:
counts[ent.entity_type] = counts.get(ent.entity_type, 0) + 1
redacted = self.redact_text(text, deduped, redact_types=redact_types)
return UnifiedAnalysisReport(
text=text,
entities=deduped,
redacted_text=redacted,
entity_counts=counts,
)
@staticmethod
def redact_text(
text: str,
entities: List[DetectedEntitySpan],
redact_types: Optional[Set[str]] = None,
) -> str:
target_types = redact_types if redact_types is not None else DEFAULT_REDACTABLE_ENTITY_TYPES
to_redact = [e for e in entities if e.entity_type in target_types]
if not to_redact:
return text
result = text
for ent in sorted(to_redact, key=lambda x: x.start, reverse=True):
token = ent.redaction_token or f"[{ent.entity_type}]"
result = result[: ent.start] + token + result[ent.end :]
return result
|