graph

ontology_platform/crawler_platform/app/domains/perfume/extractor.py

@@ -167,6 +167,7 @@ class PerfumeRuleBasedExtractor(GenericRuleBasedExtractor):
             brand = next((entity for entity in entities if entity.entity_type == "Brand"), None)
             for card in product_cards:
                 if brand:
+                    evidence = brand_evidence_for_product(brand, str(card["name"]), page_text) or brand.evidence_text or brand.name
                     claims.append(
                         ExtractedClaim(
                             str(card["name"]),
@@ -174,7 +175,7 @@ class PerfumeRuleBasedExtractor(GenericRuleBasedExtractor):
                             "hasBrand",
                             brand.name,
                             "Brand",
-                            evidence_text=brand.evidence_text or brand.name,
+                            evidence_text=evidence,
                             confidence=0.72,
                             confidence_reason="site brand inferred from listing page",
                         )
@@ -199,6 +200,7 @@ class PerfumeRuleBasedExtractor(GenericRuleBasedExtractor):
         claims: list[ExtractedClaim] = []
         brand = next((entity for entity in entities if entity.entity_type == "Brand"), None)
         if brand:
+            evidence = brand_evidence_for_product(brand, perfume.name, page_text) or brand.evidence_text or brand.name
             claims.append(
                 ExtractedClaim(
                     perfume.name,
@@ -206,7 +208,7 @@ class PerfumeRuleBasedExtractor(GenericRuleBasedExtractor):
                     "hasBrand",
                     brand.name,
                     "Brand",
-                    evidence_text=brand.evidence_text or brand.name,
+                    evidence_text=evidence,
                     confidence=0.78,
                     confidence_reason="brand pattern matched",
                 )
@@ -273,14 +275,16 @@ def extract_brand(page_text: str, product_name: str) -> str | None:
     for pattern in patterns:
         match = re.search(pattern, page_text, flags=re.IGNORECASE)
         if match:
-            return cleanup_value(match.group("brand"))
-    inferred = infer_site_brand(page_text)
+            brand = valid_brand(cleanup_value(match.group("brand")))
+            if brand:
+                return brand
+    inferred = infer_site_brand(page_text, product_name)
     if inferred:
         return inferred
     lines = [line.strip() for line in page_text.splitlines() if line.strip()]
     if len(lines) >= 2 and lines[1].lower() not in product_name.lower():
         candidate = cleanup_value(lines[1])
-        if len(candidate) <= 80 and not looks_like_navigation(candidate) and product_line_score(candidate) <= 0:
+        if valid_brand(candidate) and len(candidate) <= 80 and not looks_like_navigation(candidate) and product_line_score(candidate) <= 0:
             return candidate
     return None
 
@@ -301,6 +305,19 @@ def product_line_score(value: str) -> int:
     return score
 
 
+def looks_like_price(value: str | None) -> bool:
+    if not value:
+        return False
+    clean = cleanup_value(str(value))
+    return bool(
+        re.fullmatch(
+            r"(?:[$€£]\s*)?\d{1,3}(?:,\d{3})*(?:\.\d+)?\s*(?:원|KRW|USD|EUR|JPY|\$|€|£)?",
+            clean,
+            flags=re.IGNORECASE,
+        )
+    )
+
+
 def looks_like_metric_or_price(value: str) -> bool:
     clean = value.replace(",", "").strip()
     if re.fullmatch(r"\d+(?:\.\d+)?", clean):
@@ -385,7 +402,10 @@ def dedupe_product_cards(cards: list[dict[str, object]]) -> list[dict[str, objec
     return result
 
 
-def infer_site_brand(page_text: str) -> str | None:
+def infer_site_brand(page_text: str, product_name: str = "") -> str | None:
+    brand_context = f"{product_name}\n{page_text[:1200]}".lower()
+    if "forment" in brand_context or "포맨트" in brand_context:
+        return "FORMENT"
     if "912 공식 홈페이지" in page_text or "912" in page_text[:500]:
         return "912"
     return None
@@ -394,9 +414,19 @@ def infer_site_brand(page_text: str) -> str | None:
 def valid_brand(value: str | None) -> str | None:
     if not value or is_template_placeholder(value):
         return None
+    if looks_like_price(value) or looks_like_metric_or_price(value):
+        return None
     return value
 
 
+def brand_evidence_for_product(brand: ExtractedEntity, product_name: str, page_text: str) -> str | None:
+    if brand.name == "FORMENT" and "포맨트" in product_name:
+        return product_name
+    if brand.name.lower() in page_text.lower():
+        return brand.name
+    return None
+
+
 def extract_field_values(field: str, page_text: str) -> list[tuple[str, str]]:
     if field in NOTE_LABELS:
         return extract_labeled_values(page_text, NOTE_LABELS[field])