Week 8. 도메인 특화 프로젝트 (Case Study)

case-study/
├── ontology/
│   └── domain.owl          # 도메인 온톨로지
├── data/
│   ├── raw/                # 원본 데이터
│   └── processed/          # 전처리된 데이터
├── extraction/
│   └── pipeline.py         # 지식 추출 파이프라인
├── graph/
│   ├── schema.cypher       # Neo4j 스키마
│   └── import.py           # 데이터 임포트
├── query/
│   └── examples.cypher     # 예제 쿼리
└── app/
    └── demo.py             # 데모 앱

from owlready2 import *
 
onto = get_ontology("http://example.org/medical.owl")
 
with onto:
    class Disease(Thing): pass
    class Symptom(Thing): pass
    class Treatment(Thing): pass
    class Drug(Treatment): pass
 
    class hasSymptom(ObjectProperty):
        domain = [Disease]
        range = [Symptom]
 
    class treatedBy(ObjectProperty):
        domain = [Disease]
        range = [Treatment]
 
    class contraindicatedFor(ObjectProperty):
        domain = [Drug]
        range = [Disease]

# 증상-질병 관계 추출
def extract_disease_symptoms(medical_text):
    prompt = """
    다음 의료 텍스트에서 (질병, 증상) 쌍을 추출하세요.
 
    텍스트: {text}
 
    JSON 형식: [{{"disease": "", "symptom": ""}}]
    """
    return llm.invoke(prompt.format(text=medical_text))
 
# Neo4j 적재
def load_to_neo4j(disease, symptom):
    session.run("""
        MERGE (d:Disease {name: $disease})
        MERGE (s:Symptom {name: $symptom})
        MERGE (d)-[:HAS_SYMPTOM]->(s)
    """, disease=disease, symptom=symptom)

// 증상으로 가능한 질병 추론
MATCH (s:Symptom)<-[:HAS_SYMPTOM]-(d:Disease)
WHERE s.name IN $symptoms
WITH d, count(s) AS matchCount
ORDER BY matchCount DESC
RETURN d.name, matchCount
 
// 금기 약물 체크
MATCH (drug:Drug)-[:CONTRAINDICATED_FOR]->(condition:Disease)
WHERE condition.name IN $patient_conditions
RETURN drug.name, condition.name AS contraindication

(:Disease {name: "당뇨병", icd10: "E11"})
(:Doctor {name: "김의사", specialty: "내분비내과"})
(:Treatment {name: "인슐린 요법"})
 
(:Doctor)-[:SPECIALIZES_IN]->(:Specialty)
(:Doctor)-[:TREATS]->(:Disease)
(:Disease)-[:TREATED_BY]->(:Treatment)