File size: 12,907 Bytes
dcc24f8 |
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 |
"""
FinEE Extractor - Main orchestrator for the extraction pipeline.
Implements the 5-tier additive extraction pipeline:
- Tier 0: Hash Cache
- Tier 1: Regex Engine
- Tier 2: Rule-Based Mapping
- Tier 3: LLM (targeted extraction)
- Tier 4: Validation + Normalization
"""
import time
import logging
from typing import Optional, List, Dict, Any
from .schema import (
ExtractionResult, ExtractionConfig, TransactionType,
Category, Confidence, ExtractionSource, FieldMeta
)
from .cache import LRUCache, get_cache
from .regex_engine import RegexEngine, get_regex_engine
from .merchants import get_merchant_and_category
from .normalizer import normalize_amount, normalize_date, normalize_vpa
from .validator import repair_llm_json, validate_extraction_result
from .confidence import update_result_confidence, should_use_llm
from .prompt import get_targeted_prompt, get_full_extraction_prompt, parse_targeted_response
from .backends import get_backend, get_available_backends, BaseBackend
logger = logging.getLogger(__name__)
class FinEE:
"""
Finance Entity Extractor - Main extraction class.
Orchestrates the 5-tier additive extraction pipeline with graceful degradation.
Always returns a result, never crashes.
"""
def __init__(self, config: Optional[ExtractionConfig] = None):
"""
Initialize the extractor.
Args:
config: Extraction configuration (uses defaults if None)
"""
self.config = config or ExtractionConfig()
# Initialize components
self._cache: Optional[LRUCache] = None
self._regex_engine: Optional[RegexEngine] = None
self._backend: Optional[BaseBackend] = None
self._backend_loaded = False
# Initialize cache if enabled
if self.config.cache_enabled:
self._cache = get_cache(self.config.cache_max_size)
# Initialize regex engine
self._regex_engine = get_regex_engine()
def _lazy_load_backend(self) -> bool:
"""
Lazy load LLM backend.
Returns:
True if backend is available
"""
if self._backend_loaded:
return self._backend is not None
self._backend_loaded = True
if not self.config.use_llm:
return False
try:
self._backend = get_backend(model_id=self.config.model_id)
if self._backend:
logger.info(f"Backend loaded: {self._backend.name}")
return True
except Exception as e:
logger.warning(f"Failed to load LLM backend: {e}")
return False
def extract(self, text: str) -> ExtractionResult:
"""
Extract financial entities from text.
This is the main entry point. It runs the full 5-tier pipeline
with graceful degradation.
Args:
text: Transaction text (bank SMS, email, etc.)
Returns:
ExtractionResult with extracted entities
"""
start_time = time.time()
# Tier 0: Cache Check
if self._cache:
cached = self._cache.get(text)
if cached:
cached.processing_time_ms = (time.time() - start_time) * 1000
return cached
# Tier 1: Regex Extraction
result = self._tier1_regex(text)
# Tier 2: Rule-Based Mapping
result = self._tier2_rules(result)
# Tier 3: LLM (if needed and available)
missing_fields = result.get_missing_fields(
self.config.required_fields,
self.config.desired_fields
)
if missing_fields and self.config.use_llm:
result = self._tier3_llm(text, result, missing_fields)
# Tier 4: Validation + Normalization
result = self._tier4_validate(result)
# Calculate processing time
result.processing_time_ms = (time.time() - start_time) * 1000
# Store in cache
if self._cache and result.is_complete():
self._cache.set(text, result)
return result
def _tier1_regex(self, text: str) -> ExtractionResult:
"""
Tier 1: Extract entities using regex patterns.
Args:
text: Input text
Returns:
ExtractionResult with regex-extracted fields
"""
try:
result = self._regex_engine.extract(text)
result.raw_input = text
return result
except Exception as e:
logger.warning(f"Tier 1 (regex) failed: {e}")
return ExtractionResult(raw_input=text)
def _tier2_rules(self, result: ExtractionResult) -> ExtractionResult:
"""
Tier 2: Enrich with rule-based mappings.
Args:
result: Current extraction result
Returns:
Enriched ExtractionResult
"""
try:
# Get merchant and category from VPA
merchant, category = get_merchant_and_category(
vpa=result.vpa,
text=result.raw_input
)
if merchant and not result.merchant:
result.merchant = merchant
result.meta['merchant'] = FieldMeta(
source=ExtractionSource.RULES,
confidence=0.85
)
if category and not result.category:
result.category = Category(category) if category in [c.value for c in Category] else Category.OTHER
result.meta['category'] = FieldMeta(
source=ExtractionSource.RULES,
confidence=0.80
)
return result
except Exception as e:
logger.warning(f"Tier 2 (rules) failed: {e}")
return result
def _tier3_llm(self, text: str, result: ExtractionResult,
missing_fields: List[str]) -> ExtractionResult:
"""
Tier 3: Fill missing fields using LLM.
Uses targeted prompts for specific fields rather than full extraction.
Args:
text: Original input text
result: Current extraction result
missing_fields: Fields to extract with LLM
Returns:
Updated ExtractionResult
"""
if not self._lazy_load_backend():
logger.debug("No LLM backend available, skipping Tier 3")
return result
try:
# Load model if not already loaded
if not self._backend.is_loaded:
self._backend.load_model(self.config.model_path)
# Use targeted prompts for specific fields
for field in missing_fields:
if field in ['merchant', 'category', 'date', 'reference']:
value = self._extract_single_field(text, field)
if value:
self._set_field(result, field, value, ExtractionSource.LLM)
# If still missing critical fields, try full extraction
still_missing = result.get_missing_fields(self.config.required_fields, [])
if still_missing:
llm_result = self._full_llm_extraction(text)
if llm_result:
result.merge(llm_result, overwrite=False)
return result
except Exception as e:
logger.warning(f"Tier 3 (LLM) failed: {e}")
return result
def _extract_single_field(self, text: str, field: str) -> Optional[str]:
"""Extract a single field using targeted prompt."""
try:
prompt = get_targeted_prompt(field, text)
response = self._backend.generate(
prompt,
max_tokens=50,
temperature=self.config.llm_temperature
)
return parse_targeted_response(field, response)
except Exception as e:
logger.debug(f"Single field extraction failed for {field}: {e}")
return None
def _full_llm_extraction(self, text: str) -> Optional[ExtractionResult]:
"""Run full LLM extraction as fallback."""
try:
prompt = get_full_extraction_prompt(text)
response = self._backend.generate(
prompt,
max_tokens=self.config.llm_max_tokens,
temperature=self.config.llm_temperature
)
parsed = repair_llm_json(response)
if parsed:
result = validate_extraction_result(parsed)
result.raw_llm_output = response
# Mark all fields as LLM-sourced
for field in ['amount', 'type', 'date', 'account', 'reference',
'vpa', 'merchant', 'category']:
if getattr(result, field, None) is not None:
result.meta[field] = FieldMeta(
source=ExtractionSource.LLM,
confidence=0.70
)
return result
except Exception as e:
logger.debug(f"Full LLM extraction failed: {e}")
return None
def _set_field(self, result: ExtractionResult, field: str,
value: Any, source: ExtractionSource) -> None:
"""Set a field on the result with metadata."""
if field == 'category':
try:
value = Category(value.lower())
except (ValueError, AttributeError):
value = Category.OTHER
elif field == 'date':
value = normalize_date(value)
setattr(result, field, value)
result.meta[field] = FieldMeta(
source=source,
confidence=0.70 if source == ExtractionSource.LLM else 0.85,
raw_value=str(value)
)
def _tier4_validate(self, result: ExtractionResult) -> ExtractionResult:
"""
Tier 4: Validate and normalize all fields.
Args:
result: Extraction result to validate
Returns:
Validated and normalized result
"""
try:
# Normalize amount
if result.amount is not None:
result.amount = normalize_amount(result.amount)
# Normalize date
if result.date:
result.date = normalize_date(result.date)
# Normalize VPA
if result.vpa:
result.vpa = normalize_vpa(result.vpa)
# Update confidence
result = update_result_confidence(
result,
self.config.high_confidence_threshold,
self.config.medium_confidence_threshold
)
return result
except Exception as e:
logger.warning(f"Tier 4 (validation) failed: {e}")
result.confidence = Confidence.LOW
return result
def extract_batch(self, texts: List[str]) -> List[ExtractionResult]:
"""
Extract entities from multiple texts.
Args:
texts: List of transaction texts
Returns:
List of ExtractionResults
"""
return [self.extract(text) for text in texts]
def get_stats(self) -> Dict[str, Any]:
"""Get extraction statistics."""
stats = {
'cache_enabled': self.config.cache_enabled,
'llm_enabled': self.config.use_llm,
'available_backends': get_available_backends(),
'active_backend': self._backend.name if self._backend else None,
}
if self._cache:
cache_stats = self._cache.get_stats()
stats['cache'] = cache_stats.to_dict()
return stats
# Module-level singleton
_extractor: Optional[FinEE] = None
def get_extractor(config: Optional[ExtractionConfig] = None) -> FinEE:
"""Get or create the global extractor instance."""
global _extractor
if _extractor is None or config is not None:
_extractor = FinEE(config)
return _extractor
def extract(text: str) -> ExtractionResult:
"""
Extract financial entities from text.
Convenience function that uses the global extractor.
Args:
text: Transaction text
Returns:
ExtractionResult
"""
return get_extractor().extract(text)
|