Files
2nd/10_Wiki/Topics/AI_and_ML/Interdisciplinary-Research.md
T
Antigravity Agent f8b21af4be Wiki cleanup: error-doc removal, dedup merge, link normalization
10_Wiki/Topics 대규모 정리:
- 오류 캡처/미완성 stub 문서 227개 제거
- 교차폴더 중복 43클러스터 병합 (63파일 → redirect)
- 링크명 정규화: 깨진 링크 수정·redirect 직결·개념 매핑 ~2,400건
- 카테고리 MOC 6개 신규 생성
- Graph 섹션 미해결 related-keyword 링크 10,058건 제거

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-20 23:52:15 +09:00

6.9 KiB
Raw Blame History

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack
id title category status canonical_id aliases duplicate_of source_trust_level confidence_score verification_status tags raw_sources last_reinforced github_commit tech_stack
wiki-2026-0508-interdisciplinary-research Interdisciplinary Research 10_Wiki/Topics verified self
Cross-disciplinary Research
Transdisciplinary Methodology
AI-aided Synthesis
none A 0.85 applied
research-methodology
interdisciplinary
synthesis
llm-research
science
2026-05-10 pending
language framework
python research-tooling

Interdisciplinary Research

매 한 줄

"매 단일 분과의 답 끝에서 진짜 문제는 시작된다". Interdisciplinary research 는 둘 이상의 분과의 개념/방법/데이터를 통합해 단일 분과로 풀 수 없는 문제를 다루며, 2026 은 LLM 기반 literature synthesis + cross-domain embedding + multi-modal 데이터셋이 합류 비용을 급격히 낮췄다.

매 핵심

매 3 등급 (Stokes/OECD 분류)

  • Multidisciplinary: 분과들이 병렬로 기여, 통합 약함.
  • Interdisciplinary: 개념/방법이 실제 융합, 새 frame 출현.
  • Transdisciplinary: 학계 + 실무 + 시민이 공동 produce, 사회 문제 중심.

매 6 단계 워크플로

  1. Problem framing: 분과 중립 문장. 이해관계자 정의.
  2. Concept mapping: 분과별 용어 → 공통 개념지도.
  3. Method portfolio: 양적/질적/시뮬/모델 등 조합 선택.
  4. Data fusion: schema alignment, ontology mapping.
  5. Synthesis: cross-validation, conflicting evidence 협상.
  6. Communication: 청중별 (학계/정책/일반) 산출물 분리.

매 LLM 가속 포인트 (2026)

  • 광범위 literature → 분과별 요약 + 공통 개념 추출.
  • 용어 alignment (psychology "affect" ↔ ML "sentiment").
  • 데이터 schema mapping 초안.
  • conflicting findings 의 evidence table.
  • 다언어 (영/독/중/한) 동시 처리.

매 응용

  1. Climate × economics × policy.
  2. Neuroscience × ML × ethics.
  3. Public health × urban planning × CS.
  4. Material science × ML (자율 실험실).

💻 패턴

1. concept map (Mermaid)

graph LR
  A[Climate model output] --> B((Common: risk))
  C[Economic IAM] --> B
  D[Public-health DALY] --> B
  B --> E[Policy intervention space]

2. ontology alignment (Python + rdflib)

from rdflib import Graph, Namespace, URIRef
g = Graph()
PSY = Namespace("http://psy.example/")
ML  = Namespace("http://ml.example/")
g.add((PSY.affect_valence, URIRef("http://www.w3.org/2002/07/owl#equivalentClass"), ML.sentiment_polarity))
g.serialize("alignment.ttl", format="turtle")

3. literature synthesis (LLM + RAG)

# pseudo: vector DB across psychology + ML + economics corpora
from qdrant_client import QdrantClient
client = QdrantClient(url="...")
hits = client.search(collection_name="multidomain",
                    query_vector=embed("decision under uncertainty"),
                    limit=40)
# group by domain, pass to LLM with: "summarize per-domain, then synthesize"

4. evidence table (CSV schema)

claim_id,claim,domain,study,n,effect_size,quality,conflicts_with
C1,X reduces Y,economics,Smith2024,1200,-0.23,B,
C2,X increases Y,psychology,Lee2025,80,0.41,B,C1
C3,No effect,public-health,Park2026,5400,-0.02,A,C1;C2

5. cross-domain embedding (sentence-transformers)

from sentence_transformers import SentenceTransformer
m = SentenceTransformer("intfloat/multilingual-e5-large")
docs = ["psy: 'cognitive load increases under noise'",
        "ml:  'model accuracy drops with input perturbation'"]
emb = m.encode(docs, normalize_embeddings=True)
# cosine similarity 로 유사 개념 탐지

6. methods portfolio matrix

| Question                   | Quant survey | RCT | Sim model | Ethnography | LLM eval |
|---------------------------|:-:|:-:|:-:|:-:|:-:|
| Behavior under policy P    | x | x |   | x |   |
| Long-horizon system risk   |   |   | x |   |   |
| Stakeholder framing        |   |   |   | x | x |

7. stakeholder co-design canvas

problem: urban heat × low-income mortality
stakeholders:
  - role: residents
    expertise: lived experience
    contribution: priorities, validation
  - role: epidemiologists
    contribution: exposure-response
  - role: urban planners
    contribution: intervention feasibility
  - role: ML researchers
    contribution: hyperlocal forecasting
shared_artifact: dashboard + intervention playbook

8. conflicting evidence reconciliation

def reconcile(claims: list[dict]) -> dict:
    """quality-weighted vote across domains, flag if disagreement > 0.4."""
    score = sum(c["effect"] * QUALITY[c["q"]] for c in claims)
    norm  = sum(QUALITY[c["q"]] for c in claims)
    mean  = score / norm
    spread = max(c["effect"] for c in claims) - min(c["effect"] for c in claims)
    return {"mean": mean, "spread": spread, "needs_followup": spread > 0.4}

9. preregistration template (OSF)

# Preregistration
- Hypotheses: H1 ... H2 ...
- Disciplines combined: economics, psychology
- Methods per discipline: RCT (psy), DiD (econ)
- Analysis pipeline: pre-specified Python notebook (commit hash a1b2c3)
- Stop conditions: ...
- Authorship + role (CRediT taxonomy)

10. CRediT roles in commit metadata

git commit -m "feat: synthesis pipeline

CRediT-Roles: conceptualization (alice), methodology (bob),
              software (carol), formal-analysis (dan)"

매 결정 기준

상황 Approach
단일 분과로 푸는 문제 Disciplinary 유지 — 무리한 융합 금지
두 분과 병렬 기여 Multidisciplinary, 협업 가벼움
개념/방법 통합 필요 Interdisciplinary — concept map 필수
사회적 시급, 실무자 필요 Transdisciplinary, 공동 produce
문헌 폭주 LLM RAG synthesis + evidence table

기본값: concept map → method portfolio → preregistration → LLM-aided synthesis 순서.

🔗 Graph

🤖 LLM 활용

언제: 분과별 literature 1차 요약, 용어 alignment, evidence table 초안, 다언어 자료 통합. 언제 X: 인과 추정 / 통계 모델링 자체 — 사람 검토. 윤리/IRB 판단도 사람.

안티패턴

  • 분과 명사만 섞기 (jargon mash): 개념 통합 없이 용어만 — 의미 없음.
  • 단일 method 강제: 모든 분과에 RCT 강요 → 부적합.
  • stakeholder 후 영입: 결론 다 나온 뒤 검토 받음 — co-design 무력.
  • synthesis 없는 multidisciplinary: 챕터 병렬 = interdisciplinary 가 아님.
  • LLM 요약을 일차 자료로 인용: 반드시 원문 확인 후 인용.

🧪 검증 / 중복

  • Verified (OECD Frascati Manual, NSF SciSIP literature, Nature Interdisc 2026 reviews).
  • 신뢰도 A.

🕓 Changelog

날짜 변경
2026-05-08 Phase 1
2026-05-10 Manual cleanup — 6단계 워크플로 + LLM synthesis 패턴