[V10.346] 補強 PChome identity anchor scorer

services/marketplace_product_matcher.py

@@ -114,6 +114,8 @@ GENERIC_TOKENS = {
 SEARCH_NOISE_PHRASES = (
     "新品上市",
     "全新上市",
+    "國際航空版",
+    "超取免運",
     "任選一款",
     "任選1款",
     "任選一色",
@@ -134,6 +136,10 @@ SEARCH_NOISE_PHRASES = (
     "卸除髒汙",
     "卸除防曬",
     "卸防曬",
+    "防水眼線",
+    "寶寶牙刷",
+    "紗布牙刷",
+    "調節亮度",
     "韓國彩妝",
     "水光感",
     "官方直營",
@@ -161,6 +167,14 @@ SEARCH_NOISE_TOKENS = {
     "防曬",
     "彩妝",
     "水光感",
+    "超取",
+    "免運",
+    "航空版",
+    "國際版",
+    "附燈泡",
+    "定時",
+    "眼妝",
+    "滅菌",
     "保濕",
     "抗老",
     "超品日",
@@ -178,6 +192,17 @@ SEARCH_IDENTITY_ANCHORS = (
     "青春敷面膜",
     "長效潤膚霜",
     "小黑瓶",
+    "私密處護潔露",
+    "私密護潔露",
+    "口腔清潔棒",
+    "含氟防蛀修護牙膏",
+    "自然遮瑕素顏霜",
+    "超持久細滑眼線筆",
+    "香氛融蠟燈",
+    "水晶香氛能量寶盒禮盒組",
+    "零粉感超持久柔焦蜜粉餅",
+    "私密肌潔淨露",
+    "身體除毛器",
     "免用水潔淨液",
     "身體按摩精油",
     "按摩精油",
@@ -487,6 +512,11 @@ def _leading_brand_tokens(original: str, normalized: str) -> set[str]:
                     tokens.add(token)
 
     leading = normalized[:48]
+    leading_tokens = _tokenize(leading)
+    if leading_tokens:
+        first_token = leading_tokens[0]
+        if re.fullmatch(r"[\u4e00-\u9fff]{2,6}", first_token) and first_token not in GENERIC_TOKENS:
+            tokens.add(first_token)
     for token in _tokenize(leading):
         if re.fullmatch(r"[a-z][a-z0-9\-']{2,}", token):
             tokens.add(token)
@@ -1183,6 +1213,17 @@ def score_marketplace_match(
     ):
         score += 0.025
         reasons.append("strong_exact_spec_match")
+    shared_anchor = _shared_identity_anchor(left, right)
+    if (
+        shared_anchor
+        and brand_score >= 0.95
+        and not hard_veto
+        and price_penalty == 0
+        and spec_score >= 0.85
+        and (token_score >= 0.43 or sequence_score >= 0.58)
+    ):
+        score += 0.03
+        reasons.append("shared_identity_anchor")
     if (
         brand_score >= 0.95
         and not hard_veto
@@ -1267,6 +1308,25 @@ def _extract_anchor_phrases(token: str) -> list[str]:
     return phrases
 
 
+def _shared_identity_anchor(left: ProductIdentity, right: ProductIdentity) -> str:
+    left_anchors: set[str] = set()
+    right_anchors: set[str] = set()
+    for token in left.core_tokens:
+        left_anchors.update(_extract_anchor_phrases(token))
+    for token in right.core_tokens:
+        right_anchors.update(_extract_anchor_phrases(token))
+
+    shared = sorted(
+        {
+            anchor
+            for anchor in left_anchors & right_anchors
+            if len(anchor.replace(" ", "")) >= 5 and anchor not in SEARCH_BROAD_ANCHORS
+        },
+        key=lambda anchor: (-len(anchor.replace(" ", "")), anchor),
+    )
+    return shared[0] if shared else ""
+
+
 def _search_core_score(token: str, all_tokens: set[str]) -> tuple[int, int, str]:
     cleaned = _clean_search_phrase(token)
     if not cleaned: