Knowledge Graph 25 Classes (Mitsukoshi)

Neptune (~550K edges). VIP + 외국인 회원 + 입점 700 브랜드 + 면세 특성 반영.

---

1. 25 클래스

그룹	클래스
고객/회원	Customer · Membership(skm-member) · Foreigner (여권·국적·환율) · Persona · Segment
상품/카탈로그	Product · Brand (입점 700+ LV/Hermes/Chanel/...) · Boutique (점포 내 매장) · Category · Bundle
거래/행동	OrderTransaction (자사몰) · POSTransaction (점포 매장) · TaxRefundTransaction (면세) · CartEvent · ReviewRating
채널/캠페인	Store (19 점포) · Campaign (週年慶/SS/FW) · Promotion · Touchpoint (SMS/카톡/소홍서) · Coupon
운영/외부	SocialSignal (소홍서·Dcard·인스타) · WeatherSignal · EconomicSignal (환율) · TourismSignal (대만 관광청) · Compliance

---

2. Mitsukoshi 특화 클래스

2.1 Foreigner (외국인 관광객)

속성
foreigner_id (마스킹) · 여권 국적 (JP/HK/SG/MY/...) · 입국일 · TaxRefund 신청 여부 · 환율 적용

2.2 Boutique (점포 내 매장)

속성
boutique_id · brand_id · store_id · floor (1F/2F/...) · 면적 · 임대 조건

2.3 TaxRefundTransaction (면세 거래)

속성
txn_id · foreigner_id · txn_amount (원화/대만달러/일본엔) · refund_amount · refund_method

2.4 TourismSignal (관광 시그널)

속성
date · 입국 외국인 수 (국적별) · 평균 체류 일수 · 평균 소비

---

3. 핵심 관계

엣지 추정:

• Customer × POSTransaction (~200K)
• Boutique × Brand × Store (~5K)
• Foreigner × TaxRefundTransaction (~80K)
• Brand × Product (~100K)
• 외부 시그널 (~50K)
• 기타 (~120K)

→ 약 550K edges

---

4. openCypher 예시

4.1 S9-M — 일본인 30대 여성 면세 추천

MATCH (f:Foreigner {nationality: 'JP'})-[:CLASSIFIED_AS]->(p:Persona {age_band: '30s', gender: 'F'})
MATCH (f)-[:MADE]->(t:TaxRefundTransaction)-[:CONTAINS]->(:Product)-[:OF_BRAND]->(b:Brand)
RETURN b.name, count(t) AS popularity ORDER BY popularity DESC LIMIT 10

4.2 S10-M — 信義점 1F 럭셔리 SOV

MATCH (s:Store {name: '台北信義'})<-[:IN_STORE]-(bq:Boutique {floor: '1F'})
      -[:OF_BRAND]->(b:Brand)
MATCH (bq)<-[:AT_BOUTIQUE]-(t:POSTransaction)
WHERE t.paid_at > datetime() - duration('P30D')
RETURN b.name, sum(t.total_amount) AS gmv
ORDER BY gmv DESC

4.3 S11-M — Black VIP 회원 프리세일 추천

MATCH (c:Customer)-[:HAS_MEMBERSHIP]->(m:Membership {grade: 'Black'})
MATCH (c)-[:MADE]->(:POSTransaction)-[:CONTAINS]->(p:Product)
      -[:OF_BRAND]->(b:Brand)
WITH c, b, count(*) AS frequency
ORDER BY frequency DESC LIMIT 5
RETURN c.customer_id, collect(b.name)[..5] AS preferred_brands

---

5. cohort_tag

값	의미
real	PII 마스킹 회원 (N=2K Customer + 500 Foreigner)
synth	합성 49.5K
external	소셜·환율·관광·기상

---

6. OpenSearch 인덱스

인덱스	분석기
idx_product	Smartcn (번체중) + Standard
idx_customer	Smartcn
idx_review (소홍서/Dcard 외부)	Smartcn + Kuromoji (일본어)
idx_brand	Standard (브랜드명 + 메타)