"""Document type classifier using keyword heuristics.""" from __future__ import annotations import re from typing import Tuple _INVOICE_KEYWORDS = [ r"\binvoice\b", r"\binv[-\s]?\d", r"\bbill\s+to\b", r"\bdue\s+date\b", r"\btax\s+invoice\b", r"\bgstin\b", r"\bremit\s+to\b", r"\bpayment\s+due\b", r"\binvoice\s+number\b", r"\binvoice\s+no\b", r"\binvoice\s+date\b", ] _RECEIPT_KEYWORDS = [ r"\breceipt\b", r"\bcashier\b", r"\bamount\s+paid\b", r"\bpayment\s+received\b", r"\bthank\s+you\s+for\b", r"\bchange\b", r"\bcash\s+tender\b", r"\breceived\s+from\b", r"\bpos\s+terminal\b", r"\btransaction\s+id\b", ] _PO_KEYWORDS = [ r"\bpurchase\s+order\b", r"\bpo\s+number\b", r"\bpo\s+no\b", r"\bship\s+to\b", r"\border\s+date\b", r"\bdelivery\s+date\b", r"\bvendor\b", r"\bbuyer\b", r"\bprocurement\b", r"\brequisition\b", ] def _count_matches(text: str, patterns: list) -> int: count = 0 t = text.lower() for pattern in patterns: if re.search(pattern, t): count += 1 return count def classify_document(text: str) -> Tuple[str, float]: """ Classify document type from extracted text. Returns (document_type, confidence) where confidence is 0.0 to 1.0. """ invoice_score = _count_matches(text, _INVOICE_KEYWORDS) receipt_score = _count_matches(text, _RECEIPT_KEYWORDS) po_score = _count_matches(text, _PO_KEYWORDS) scores = { "invoice": invoice_score, "receipt": receipt_score, "purchase_order": po_score, } max_score = max(scores.values()) if max_score == 0: return "unknown", 0.0 best = max(scores, key=scores.get) total = sum(scores.values()) confidence = max_score / total if total > 0 else 0.0 if max_score >= 3: confidence = min(0.95, confidence + 0.3) elif max_score >= 2: confidence = min(0.85, confidence + 0.15) elif max_score == 1: confidence = min(0.65, confidence) return best, round(confidence, 4)