ontology

crawler_platform/app/domains/perfume/extractor.py

@@ -118,7 +118,7 @@ class PerfumeRuleBasedExtractor(GenericRuleBasedExtractor):
         price = find_price(page_text)
         if price:
             attrs["price"] = {k: v for k, v in price.items() if k != "evidence"}
-        entities = [ExtractedEntity("Perfume", product_name, attrs, confidence=0.68)]
+        entities = [ExtractedEntity("Perfume", product_name, attrs, confidence=0.74)]
         if brand:
             entities.append(ExtractedEntity("Brand", brand, confidence=0.62))
         for field, entity_type in [
@@ -187,7 +187,7 @@ class PerfumeRuleBasedExtractor(GenericRuleBasedExtractor):
                             "hasPrice",
                             object_value=card["price"],
                             evidence_text=str(card.get("evidence") or card["name"]),
-                            confidence=0.76,
+                            confidence=0.84,
                             confidence_reason="Korean product listing price pattern matched",
                         )
                     )
@@ -236,7 +236,7 @@ class PerfumeRuleBasedExtractor(GenericRuleBasedExtractor):
                     "hasPrice",
                     object_value={k: v for k, v in price.items() if k != "evidence"},
                     evidence_text=price["evidence"],
-                    confidence=0.7,
+                    confidence=0.86,
                     confidence_reason="price pattern matched",
                 )
             )
@@ -249,10 +249,19 @@ class PerfumeRuleBasedExtractor(GenericRuleBasedExtractor):
 
 
 def extract_product_name(page_text: str) -> str:
-    for line in page_text.splitlines()[:8]:
+    candidates: list[tuple[int, str]] = []
+    for line in page_text.splitlines()[:12]:
         clean = line.strip()
-        if clean and not looks_like_navigation(clean) and not is_template_placeholder(clean):
-            return clean[:240]
+        if not clean or looks_like_navigation(clean) or is_template_placeholder(clean):
+            continue
+        if looks_like_metric_or_price(clean):
+            continue
+        candidates.append((product_line_score(clean), clean[:240]))
+    strong = [candidate for candidate in candidates if candidate[0] > 0]
+    if strong:
+        return max(strong, key=lambda item: item[0])[1]
+    if candidates:
+        return candidates[0][1]
     return first_non_empty_line(page_text) or "Unknown Perfume"
 
 
@@ -265,14 +274,40 @@ def extract_brand(page_text: str, product_name: str) -> str | None:
         match = re.search(pattern, page_text, flags=re.IGNORECASE)
         if match:
             return cleanup_value(match.group("brand"))
+    inferred = infer_site_brand(page_text)
+    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):
+        if len(candidate) <= 80 and not looks_like_navigation(candidate) and product_line_score(candidate) <= 0:
             return candidate
     return None
 
 
+def product_line_score(value: str) -> int:
+    lower = value.lower()
+    score = 0
+    if re.search(r"\d+\s*(?:ml|g|개입)", lower):
+        score += 4
+    if any(keyword in value for keyword in ["향수", "디퓨저", "스프레이", "핸드크림", "미스트", "샤쉐", "퍼퓸"]):
+        score += 3
+    if value.startswith("[") or any(keyword in value for keyword in ["기획", "추가할인", "모음"]):
+        score += 2
+    if "912" in value:
+        score += 2
+    if "시작" in value and not re.search(r"\d+\s*(?:ml|g|개입)", lower):
+        score -= 4
+    return score
+
+
+def looks_like_metric_or_price(value: str) -> bool:
+    clean = value.replace(",", "").strip()
+    if re.fullmatch(r"\d+(?:\.\d+)?", clean):
+        return True
+    return bool(re.fullmatch(r"\d+(?:\.\d+)?\s*(?:원|krw|usd)?", clean, flags=re.IGNORECASE))
+
+
 def extract_product_cards(page_text: str) -> list[dict[str, object]]:
     lines = [line.strip() for line in page_text.splitlines() if line.strip()]
     cards: list[dict[str, object]] = []