V10.398 recover true low confidence matches

2026-05-24 12:05:30 +08:00
parent 8cc197aba2
commit 4622be3441
4 changed files with 146 additions and 2 deletions
--- a/config.py
+++ b/config.py
@@ -325,7 +325,7 @@ YOUTUBE_API_KEY = os.getenv('YOUTUBE_API_KEY', '')
 # ==========================================
 # 系統版本與路徑
 # ==========================================
-SYSTEM_VERSION = "V10.397"
+SYSTEM_VERSION = "V10.398"
 LOG_FILE_PATH = os.path.join(BASE_DIR, 'logs/system.log')
 public_url = PUBLIC_URL  # 用於模板顯示

--- a/docs/memory/history_logs.md
+++ b/docs/memory/history_logs.md
@@ -12,6 +12,9 @@

 ## 📅 詳細更新日誌 (考古存檔)

+### 2026-05-24：PChome 近門檻身份回收第二輪
+- **V10.398 true low confidence 保守回收**: marketplace matcher 針對正式前段 `true_low_confidence` 補一輪 focused exact identity lines，讓 Baan 嬰兒修護唇膏、植村秀 3D 極細防水眼線膠筆、YSL 恆久完美透膚煙染腮紅、HH 私密植萃美白緊緻凝露、Lab52 學習刷牙漱口水、Benefit 經典菲菲染唇液、Herb24 晨霧純精油擴香儀、Pavaruni 40 香味 10ml 精油與 GATSBY 爆水擦澡濕巾等近門檻真同款可被回收；未放寬 `MIN_MATCH_SCORE`。同版保留 peripera 多色任選對單一色號、LUNASOL 頰彩對眼彩組、MUJI 細軸棉棒對黑色棉棒的低信心保護，並讓多組件套組即使達強身份證據也停在 `identity_review`，避免總價被誤當精準價格告警。
+
 ### 2026-05-21：瀏覽器測試守門與 PChome 熱路徑優化
 - **V10.397 離線 audit false negative 收斂**: marketplace matcher 針對 audit 剩餘 fresh bad 補三個保守修正：`200ml+200ml` 這類括號內規格串不再被計成額外組件；Kiehl’s `1號護唇膏` 的 `1` 視為商品線名稱而非色號；P.SHINE BEAUTY FOOT 雙面足部去硬皮磨砂棒以品牌 + 足部硬皮磨砂棒語意進 `identity_review`。同批補 focused exact identity lines，讓 Biodance、SAB、LUSH、Kanebo、ARTMIS、Nailmatic、小浪、YUNMI、AQUIESSE、資生堂等低分但強證據同款被正確拉回；高證據 exact 才可進 `price_alert_exact`，證據不足者仍進 `identity_review`。離線 audit fresh bad 從 9 降到 6，剩餘皆為多組件/套組差異。
 - **V10.396 多選 catalog 對 generic count 組合放行**: marketplace matcher 對「多款任選 catalog listing」對上同數量 generic `N入組` 候選新增保守豁免：需品牌、品類、基礎規格與數量一致，且 generic 端沒有具名色款/香味選項，才不觸發 `variant_option_conflict`。John’s Blend 香氛擴香罐 85g 任選 3 入 vs PChome 3入組會進 `identity_review`，不直接價格告警。
--- a/services/marketplace_product_matcher.py
+++ b/services/marketplace_product_matcher.py
@@ -352,6 +352,16 @@ SEARCH_IDENTITY_ANCHORS = (
    "麗駐蘭修復舒緩面膜",
    "濕度數顯智能加濕器",
    "新艷陽夏水離子熱防禦隔離露",
+    "3d極細防水眼線膠筆",
+    "恆久完美透膚煙染腮紅",
+    "私密植萃美白緊緻凝露",
+    "學習刷牙漱口水",
+    "羅馬柱智慧居家車用香氛機",
+    "經典菲菲染唇液",
+    "染唇液",
+    "晨霧純精油擴香儀",
+    "天然植物香氛精油",
+    "爆水擦澡濕巾",
    "3d立體持色眉彩盤",
    "細芯睛彩雙頭眉筆",
    "雙頭旋轉極細眉筆",
@@ -598,6 +608,7 @@ BRAND_ALIAS_OVERRIDES = {
    "jo malone": ("jo malone",),
    "prada": ("prada", "普拉達"),
    "za": ("za",),
+    "hh": ("hh", "草本新淨界"),
    "小浪": ("小浪",),
    "xiaomi": ("小米有品", "小米", "xiaomi"),
    "mac": ("m.a.c", "mac", "m a c"),
@@ -1649,6 +1660,7 @@ def _classify_match_quality(
    shared_anchor: str,
    shared_models: set[str],
    catalog_count_omission: bool,
+    multi_component_pair: bool,
 ) -> tuple[str, str, str]:
    """Map raw matcher scores into operator-facing price comparison lanes."""
    reason_set = set(reasons)
@@ -1670,6 +1682,8 @@ def _classify_match_quality(
        and (direct_spec_evidence or (shared_anchor and token_score >= 0.62 and sequence_score >= 0.58))
    )
    if strong_identity_evidence and not catalog_count_omission:
+        if multi_component_pair:
+            return "exact", "manual_review", "identity_review"
        return "exact", "total_price", "price_alert_exact"

    if score >= 0.76:
@@ -2288,6 +2302,7 @@ def score_marketplace_match(
        shared_anchor=shared_anchor,
        shared_models=shared_models,
        catalog_count_omission=catalog_count_omission,
+        multi_component_pair=_has_multi_component(left) and _has_multi_component(right),
    )
    evidence_flags = _build_evidence_flags(
        brand_score=brand_score,
@@ -2571,6 +2586,20 @@ def _has_baan_baby_lip_catalog_alignment(left: ProductIdentity, right: ProductId
    )


+def _has_pavaruni_40_scent_oil_alignment(left: ProductIdentity, right: ProductIdentity) -> bool:
+    left_text = left.searchable_name
+    right_text = right.searchable_name
+    return (
+        "pavaruni" in (left.brand_tokens & right.brand_tokens)
+        and "天然植物" in f"{left_text} {right_text}"
+        and "精油" in left_text
+        and "精油" in right_text
+        and _has_shared_volume(left, right, 10)
+        and ("40香味" in left_text or "40種香味" in left_text)
+        and ("40香味" in right_text or "40種香味" in right_text)
+    )
+
+
 def _has_shared_count(left: ProductIdentity, right: ProductIdentity, count: int, unit: str) -> bool:
    return (count, unit) in set(left.counts) and (count, unit) in set(right.counts)

@@ -2677,6 +2706,63 @@ def _has_focused_low_score_exact_identity_line(left: ProductIdentity, right: Pro
        and "隔離露" in right_text
    ):
        return "shiseido_blue_sunscreen"
+    if (
+        {"baan", "貝恩"} & (left.brand_tokens & right.brand_tokens)
+        and "嬰兒修護唇膏" in left_text
+        and "嬰兒修護唇膏" in right_text
+        and left.product_type == right.product_type == "護唇膏"
+    ):
+        return "baan_baby_lip_base_catalog"
+    if (
+        {"shu uemura", "植村秀"} & (left.brand_tokens & right.brand_tokens)
+        and "3d極細防水眼線膠筆" in left_text
+        and "3d極細防水眼線膠筆" in right_text
+    ):
+        return "shu_3d_eyeliner"
+    if (
+        {"ysl", "聖羅蘭"} & (left.brand_tokens & right.brand_tokens)
+        and "恆久完美透膚煙染腮紅" in left_text
+        and "恆久完美透膚煙染腮紅" in right_text
+    ):
+        return "ysl_blush_catalog"
+    if (
+        {"hh", "草本新淨界"} & (left.brand_tokens & right.brand_tokens)
+        and "私密植萃美白緊緻凝露" in left_text
+        and "私密植萃美白緊緻凝露" in right_text
+        and _has_shared_volume(left, right, 30)
+    ):
+        return "hh_private_gel"
+    if (
+        {"lab52", "齒妍堂"} & (left.brand_tokens & right.brand_tokens)
+        and "學習刷牙漱口水" in left_text
+        and "學習刷牙漱口水" in right_text
+        and _has_overlapping_base_spec(left, right)
+    ):
+        return "lab52_mouthwash"
+    if (
+        "benefit" in (left.brand_tokens & right.brand_tokens)
+        and "染唇液" in left_text
+        and "染唇液" in right_text
+        and "唇頰兩用" in pair_text
+    ):
+        return "benefit_lip_tint"
+    if (
+        {"herb24", "草本"} & (left.brand_tokens & right.brand_tokens)
+        and "晨霧純精油擴香儀ii" in left_text
+        and "晨霧純精油擴香儀ii" in right_text
+        and (("霧黑" in left_text and "黑色" in right_text) or ("霧黑" in right_text and "黑色" in left_text))
+    ):
+        return "herb24_mist_diffuser_black"
+    if _has_pavaruni_40_scent_oil_alignment(left, right):
+        return "pavaruni_40_scent_oil"
+    if (
+        "gatsby" in (left.brand_tokens & right.brand_tokens)
+        and "爆水擦澡濕巾" in left_text
+        and "爆水擦澡濕巾" in right_text
+        and "24張入" in left_text
+        and "24張入" in right_text
+    ):
+        return "gatsby_body_wipes_24"
    return ""


@@ -2735,6 +2821,8 @@ def _has_variant_descriptor_conflict(left: ProductIdentity, right: ProductIdenti
        return False
    if _has_baan_baby_lip_catalog_alignment(left, right):
        return False
+    if _has_pavaruni_40_scent_oil_alignment(left, right):
+        return False
    if _is_relove_private_cleanser_line(left, right):
        return False
    if (
--- a/tests/test_marketplace_product_matcher.py
+++ b/tests/test_marketplace_product_matcher.py
@@ -1013,6 +1013,51 @@ def test_marketplace_matcher_promotes_focused_low_score_exact_identity_lines():
            "《SHISEIDO 資生堂》新艷陽夏水離子熱防禦隔離露 50ml",
            "focused_exact_identity_shiseido_blue_sunscreen",
        ),
+        (
+            "【Baan 貝恩】嬰兒修護唇膏4.5g",
+            "【貝恩】嬰兒修護唇膏(原味/草莓)",
+            "focused_exact_identity_baan_baby_lip_base_catalog",
+        ),
+        (
+            "【植村秀】官方直營 3D極細防水眼線膠筆(Shu uemura)",
+            "《Shu Uemura 植村秀》3D極細防水眼線膠筆 0.08g-#深棕",
+            "focused_exact_identity_shu_3d_eyeliner",
+        ),
+        (
+            "【YSL】官方直營 恆久完美透膚煙染腮紅(任選1款/新品上市)",
+            "【YSL聖羅蘭】恆久完美透膚煙染腮紅 6g ( #12/ #57/ #93)",
+            "focused_exact_identity_ysl_blush_catalog",
+        ),
+        (
+            "【HH草本新淨界】私密植萃美白緊緻凝露30ml(私密美白 保濕緊緻 私密緊實)",
+            "HH 私密植萃美白緊緻凝露(30ml)",
+            "focused_exact_identity_hh_private_gel",
+        ),
+        (
+            "【Lab52 齒妍堂】學習刷牙漱口水170g/瓶(食品級配方/牙菌斑顯示劑/口腔清潔居家檢測)",
+            "Lab52齒妍堂 汪汪隊立大功學習刷牙漱口水170g",
+            "focused_exact_identity_lab52_mouthwash",
+        ),
+        (
+            "【benefit 貝玲妃】經典染唇液(唇頰兩用/持久不脫妝)",
+            "【Benefit】經典菲菲染唇液6ml(唇頰兩用/持久不脫妝)多色",
+            "focused_exact_identity_benefit_lip_tint",
+        ),
+        (
+            "【草本24】Herb24 晨霧純精油擴香儀II_霧黑－無須用水（贈－紅桔 純質精油）(歐盟EMC證書 主機享半年保固)",
+            "【草本24。Herb24】晨霧純精油擴香儀II_黑色（贈：任選價值780元純質精油10ml）",
+            "focused_exact_identity_herb24_mist_diffuser_black",
+        ),
+        (
+            "【Pavaruni】美國天然植物精油40香味10ml(香薰 擴香 萃取 薰香機 水氧機 薰衣草 檀香 玫瑰 雪松 白茶 茉莉)",
+            "【美國Pavaruni】天然植物香氛精油40種香味10ml 多款任選",
+            "focused_exact_identity_pavaruni_40_scent_oil",
+        ),
+        (
+            "【GATSBY】爆水擦澡濕巾24張入(涼感乾洗澡)",
+            "GATSBY 爆水擦澡濕巾24張入(240g)",
+            "focused_exact_identity_gatsby_body_wipes_24",
+        ),
    ]

    for momo_name, competitor_name, expected_reason in cases:
@@ -1033,8 +1078,16 @@ def test_marketplace_matcher_keeps_high_variant_low_score_lines_outside_focused_
        "【Lactacyd 立朵舒】清新舒涼 私密潔浴露250ml(天然薄荷涼感 私密清潔 私密保養 乾癢修護)",
        "Lactacyd 立朵舒私密潔浴露-生理呵護250ml",
    )
+    lunasol = score_marketplace_match(
+        "即期品【Kanebo 佳麗寶】LUNASOL 晶巧霓光/明艷頰彩(多色任選)",
+        "【Kanebo 佳麗寶】LUNASOL 晶巧霓光眼彩寵愛組",
+    )
+    muji_swab = score_marketplace_match(
+        "【MUJI 無印良品】細軸棉棒/200支",
+        "【MUJI 無印良品】棉棒(黑色)/200支",
+    )

-    for diagnostics in (lush, lactacyd):
+    for diagnostics in (lush, lactacyd, lunasol, muji_swab):
        assert diagnostics.score < 0.76
        assert not any(reason.startswith("focused_exact_identity_") for reason in diagnostics.reasons)