1. 연구 주제명

"블랙박스 해부: LLM과 RAG의 편향성 전이 경로 추적" (Inside the Black Box: Tracing the Neural Roots of Bias in LLM & RAG)

2. 한 줄 요약

<aside> 💡

단순 통계 분석을 넘어, 웹 데이터의 편향이 모델 내부의 어떤 뉴런과 레이어를 타고 흘러가 최종 검색(RAG) 결과까지 오염시키는지 인과적으로 규명하는 연구.

</aside>

3. 핵심 도식

image.png

4. 연구 개요 및 목표

5. 목표

6. 연구실 선행연구

  1. [WSDM2026] Jaebeom You, Seung-Kyu Hong, Ling Liu, Kisung Lee, and Hyuk-Yoon Kwon*, ****"Geo-Personalization Bias in News Search: Analyzing Filter Bubbles in Search Engine Results with Multi-Perspective LLM Annotation," In Proc. The 19th ACM International Conference on Web Search and Data Mining (WSDM), Boise, Idaho, USA, Feb. 2026. TOP CONFERENCE in Web and Data Mining BK IF=3 REGULAR Acceptance rate=16.3%
  2. [WSDM2026] Jaebeom You, Jaewon Lee, Sehun Lee, and Hyuk-Yoon Kwon*, ****"From Data to Model in Bias: A Statistical Analysis of Political Bias in the C4 Corpus and Its Impact on LLMs," In Proc. The 19th ACM International Conference on Web Search and Data Mining (WSDM), Boise, Idaho, USA, Feb. 2026. TOP CONFERENCE in Web and Data Mining BK IF=3 REGULAR Acceptance rate=16.3%
  3. [CIKM2025] Jaebeom You, Seung-Kyu Hong, Ling Liu, Kisung Lee, and Hyuk-Yoon Kwon,* "FAIR-SE: Framework for Analyzing Information Disparities in Search Engines with Diverse LLM-Generated Personas," In Proc. the 34th ACM International Conference on Information and Knowledge Management (CIKM), Seoul, Korea, Nov. 2025. TOP CONFERENCE in IR and DB BK IF=3 REGULAR