Dhurgh's picture
Initial commit
410242f
Raw
History Blame Contribute Delete
9.11 kB
"""
Data Extractor Agent - Extracts structured data from documents.
Supports invoices, forms, contracts, and general text extraction.
"""
import re
import logging
from typing import Dict, List, Optional, Any
from dataclasses import dataclass
from enum import Enum
from datetime import datetime
logger = logging.getLogger(__name__)
class FieldType(str, Enum):
r"""Supported field types"""
TEXT = r"text"
EMAIL = r"email"
PHONE = r"phone"
DATE = r"date"
AMOUNT = r"amount"
ADDRESS = r"address"
URL = r"url"
REFERENCE = r"reference"
@dataclass
class ExtractedField:
r"""Extracted field from document"""
name: str
value: Any
field_type: FieldType
confidence: float
location: Optional[Dict] = None
@dataclass
class ExtractionResult:
r"""Result from data extraction"""
extracted_fields: List[ExtractedField]
document_type: str
raw_text: str
confidence: float
structured_data: Dict[str, Any]
class DataExtractor:
r"""Extract structured data from text"""
# Regex patterns for common fields
PATTERNS = {
FieldType.EMAIL: r'\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Z|a-z]{2,}\b',
FieldType.PHONE: r'\b(\+?1[-.\s]?)?\(?[0-9]{3}\)?[-.\s]?[0-9]{3}[-.\s]?[0-9]{4}\b',
FieldType.DATE: r'\b(?:\d{1,2}[-/]\d{1,2}[-/]\d{2,4}|\d{4}[-/]\d{1,2}[-/]\d{1,2})\b',
FieldType.AMOUNT: r'\$[\d,]+\.?\d*|\b\d+\.?\d*\s*(?:USD|EUR|GBP)',
FieldType.URL: r'https?://[^\s]+',
}
# Invoice-specific extractors
INVOICE_FIELDS = {
r'invoice_number': ['invoice.*?#?(\d+)', r'invoice.*?:.*?(\w+)'],
r'invoice_date': ['invoice.*?date.*?(\d{1,2}[-/]\d{1,2}[-/]\d{2,4})'],
r'due_date': ['due.*?date.*?(\d{1,2}[-/]\d{1,2}[-/]\d{2,4})'],
r'total_amount': ['total.*?:?\s*\$?([\d,]+\.?\d*)', r'amount.*?due.*?\$?([\d,]+\.?\d*)'],
r'vendor_name': ['from:.*?([A-Za-z\s&]+)', r'vendor.*?:?\s*([A-Za-z\s&]+)'],
r'customer_name': ['bill.*?to:?\s*([A-Za-z\s&]+)', r'customer.*?:?\s*([A-Za-z\s&]+)'],
}
def __init__(self):
r"""Initialize data extractor"""
self.compiled_patterns = {
field_type: re.compile(pattern, re.IGNORECASE)
for field_type, pattern in self.PATTERNS.items()
}
def extract(
self,
text: str,
document_type: str = r"general",
custom_fields: Optional[Dict[str, str]] = None
) -> ExtractionResult:
r"""
Extract structured data from document text.
Args:
text: Document text to extract from
document_type: Type of document (invoice, form, etc.)
custom_fields: Custom field patterns to extract
Returns:
ExtractionResult with extracted fields and structured data
r"""
try:
extracted_fields = []
structured_data = {}
# Extract common fields
for field_type in FieldType:
matches = self._extract_by_type(text, field_type)
for match, confidence in matches:
field = ExtractedField(
name=f"{field_type.value}_{len(extracted_fields)}",
value=match,
field_type=field_type,
confidence=confidence
)
extracted_fields.append(field)
# Extract document-specific fields
if document_type.lower() == r"invoice":
invoice_fields = self._extract_invoice_fields(text)
extracted_fields.extend(invoice_fields)
structured_data['invoice'] = {
f.name: f.value for f in invoice_fields
}
# Extract custom fields
if custom_fields:
custom_extracted = self._extract_custom_fields(text, custom_fields)
extracted_fields.extend(custom_extracted)
structured_data['custom'] = {
f.name: f.value for f in custom_extracted
}
# Calculate overall confidence
confidences = [f.confidence for f in extracted_fields]
avg_confidence = sum(confidences) / len(confidences) if confidences else 0.0
return ExtractionResult(
extracted_fields=extracted_fields,
document_type=document_type,
raw_text=text,
confidence=avg_confidence,
structured_data=structured_data
)
except Exception as e:
logger.error(f"Extraction failed: {str(e)}")
return ExtractionResult(
extracted_fields=[],
document_type=document_type,
raw_text=text,
confidence=0.0,
structured_data={}
)
def _extract_by_type(self, text: str, field_type: FieldType) -> List[tuple]:
r"""Extract fields by type using regex"""
matches = []
pattern = self.compiled_patterns.get(field_type)
if not pattern:
return matches
for match in pattern.finditer(text):
value = match.group(0)
# Confidence based on pattern match
confidence = 0.8
matches.append((value, confidence))
return matches
def _extract_invoice_fields(self, text: str) -> List[ExtractedField]:
r"""Extract invoice-specific fields"""
fields = []
text_lower = text.lower()
for field_name, patterns in self.INVOICE_FIELDS.items():
for pattern in patterns:
try:
regex = re.compile(pattern, re.IGNORECASE)
matches = regex.findall(text_lower)
if matches:
value = matches[0]
# Clean up value
if field_name == r'total_amount':
value = value.replace(',', r'')
field = ExtractedField(
name=field_name,
value=value,
field_type=FieldType.TEXT,
confidence=0.75
)
fields.append(field)
break
except Exception as e:
logger.warning(f"Pattern extraction failed for {field_name}: {e}")
return fields
def _extract_custom_fields(
self,
text: str,
custom_fields: Dict[str, str]
) -> List[ExtractedField]:
r"""Extract custom defined fields"""
fields = []
for field_name, pattern in custom_fields.items():
try:
regex = re.compile(pattern, re.IGNORECASE | re.DOTALL)
matches = regex.findall(text)
for match in matches:
field = ExtractedField(
name=field_name,
value=match if isinstance(match, str) else match[0],
field_type=FieldType.TEXT,
confidence=0.7
)
fields.append(field)
except Exception as e:
logger.warning(f"Custom field extraction failed for {field_name}: {e}")
return fields
def extract_entities(self, text: str) -> Dict[str, List[str]]:
r"""Extract named entities (persons, organizations, locations)"""
entities = {
r'persons': [],
r'organizations': [],
r'locations': [],
r'money': [],
r'dates': [],
r'organizations': []
}
# Simple pattern-based entity extraction
# In production, use spaCy or similar NLP libraries
# Find all capitalized sequences (potential names)
name_pattern = r'\b([A-Z][a-z]+ (?:[A-Z][a-z]+)*)\b'
entities['persons'] = re.findall(name_pattern, text)
# Money amounts
money_pattern = r'\$[\d,]+\.?\d*'
entities['money'] = re.findall(money_pattern, text)
# Dates
entities['dates'] = re.findall(self.PATTERNS[FieldType.DATE], text)
return entities
def batch_extract(
self,
documents: List[Dict],
document_type: str = r"general"
) -> List[ExtractionResult]:
r"""Extract from multiple documents"""
results = []
for doc in documents:
text = doc.get('text', r'')
doc_type = doc.get('type', document_type)
custom_fields = doc.get('custom_fields')
result = self.extract(text, doc_type, custom_fields)
results.append(result)
return results