cha-interview-bot 신뢰 설계 컴포넌트 18종

박대근 교수 풀 디지털 트윈 기반 신입생 전공상담 AI 면담봇 — 신뢰 이론 매핑
v1.0 · 2026-05-05 · github.com/sungbongju/cha-interview-bot
설계 개요 — 본 봇은 신뢰 이론의 4개 층위(거시 정체성 · 콘텐츠 품질 · 인터랙션 미시 신호 · 메타 신호)에 걸쳐 18개 트러스트 컴포넌트를 의도적으로 배치하였다. 각 컴포넌트는 사용자 인지(awareness)와 신뢰 효과(effectiveness)의 두 차원에서 측정 가능하며, 본 연구는 어느 컴포넌트가 작동하고 어느 컴포넌트가 실패하는가를 검증한다. LLM은 온프레미스 Gemma4를 프롬프트 엔지니어링과 RAG로 유료 LLM 수준의 성과로 끌어올렸으며, 이는 본 연구의 system contribution이다.
L1 거시 정체성 (3) L2 콘텐츠 품질 (4) L3 인터랙션 미시 (5) L4 메타 신호 (6)

Layer 1. 거시 정체성 신호

3 / 18
봇이 "누구인가"의 정체성 정합 신호. 본 연구의 가장 강한 차별점.
01 박대근 교수 풀 디지털 트윈 Identity transfer
봇 기능
HeyGen Studio Avatar(직접 녹화) + 음성 클로닝 + 본인 Q&A RAG + 본인 말투 시스템 프롬프트의 5-차원 정체성 정합
코드 위치
App.jsx AVATAR_ID, VOICE_ID / server/cha-rag.js
의도
실재 인물에 대한 신뢰가 디지털 트윈으로 이전(identity-based trust transfer)
레퍼런스
Doney & Cannon (1997); Pataranutaporn et al. (2021)
02 기관 신원 표시 Institutional authority
봇 기능
UI 헤더·OG 메타·인사말에 "차의과학대학교 신입생 전공상담" 일관 노출
코드 위치
index.html meta tags / App.jsx getGreetingText()
의도
출처 신뢰성(source credibility)을 통한 institutional authority 신호
레퍼런스
Hovland & Weiss (1951); Bhattacherjee (2002)
03 AI 정체 명시 Honest disclosure
봇 기능
인사말과 아바타 이름판에 "AI 면담 어시스턴트" 명시
코드 위치
App.jsx getGreetingText() / AvatarPanel.jsx nameplate
의도
봇임을 알고 상호작용하도록 하여 적정 신뢰(calibrated trust) 형성
레퍼런스
Liao & Sundar (2022)

Layer 2. 콘텐츠 품질 신호

4 / 18
LLM이 "무엇을 어떻게 말하는가"의 품질. 온프레미스 Gemma4 + RAG + 프롬프트 엔지니어링이 핵심 system contribution.
04 RAG 기반 사실성 Ability — 환각 차단
봇 기능
bge-m3 임베딩, 본인 Q&A 61개 청크에서 top-5 검색, 코사인 유사도 ≥ 0.25
코드 위치
server/cha-rag.js retrieve()
의도
LLM 환각 차단으로 능력(Ability) 신뢰 신호 생성
레퍼런스
Ji et al. (2023)
05 한계의 명시적 인정 Calibrated trust
봇 기능
시스템 프롬프트에 "참고 자료에 없는 내용은 추측 금지, '교수님께 직접 여쭤보시는 걸 추천드려요'로 안내" 규칙
코드 위치
server/interview-chat.js buildSystemPrompt()
의도
정직성(Integrity) 신호이자 trust calibration의 핵심 메커니즘
레퍼런스
Lee & See (2004); Jacovi et al. (2021)
06 해요체와 따뜻한 말투 Benevolence
봇 기능
시스템 프롬프트의 "따뜻하고 친근하게, 해요체로 대화" 규칙 + 박대근 교수 본인 말투를 모사한 RAG 답변
코드 위치
server/interview-chat.js + RAG chunks
의도
선의(Benevolence)·정서적 신뢰 신호 — ABI 차원 중 의인화 단서에 가장 민감
레퍼런스
Pelau et al. (2021); Lombard & Xu (2021)
07 출력 형식 강제 (JSON · 이모지 제거) 일관성 · 전문성
봇 기능
{"reply":..., "ttsReply":...} JSON 강제 출력 + stripEmoji() 후처리
코드 위치
server/interview-chat.js buildSystemPrompt() + stripEmoji()
의도
일관된 전문성 신호이자 sycophancy(아첨성 응답) 차단
레퍼런스
Ji et al. (2023)

Layer 3. 인터랙션 미시 신호

5 / 18
평소엔 보이지 않다가 고장 시에만 가시화되는 미세 신호. A/B 실험·breakdown task 측정에 적합.
08 응답 레이턴시 관리 Turn-taking 자연성
봇 기능
Mobile/PC별 silenceMs(2000/1600ms)와 echoGuardMs(1200/2400ms) 차등 튜닝
코드 위치
App.jsx getSilenceMs(), getEchoGuardMs()
의도
STT→LLM→TTS 직렬 파이프라인에서 자연스러운 turn-taking 호흡 유지
레퍼런스
Skantze (2021); Cowan et al. (2017)
09 Echo guard 다층 방어 무한루프 차단
봇 기능
isSpeakingRef, echoGuardUntilRef, lastSubmittedSpeechRef, mergeTranscript() 등 5중 방어
코드 위치
App.jsx 다수 useEffect / sendMessage()
의도
봇 발화가 사용자 질문으로 오인지되는 무한루프 차단 — 신뢰의 보이지 않는 인프라
레퍼런스
Skantze (2021)
10 ESC 발화 인터럽트 User agency
봇 기능
window+document 양쪽 capture phase에 등록되어 봇 말 도중 즉시 중단 가능
코드 위치
App.jsx interruptAvatar() + handleGlobalKeydown
의도
사용자 통제권(user agency) 보장 — 적정 신뢰의 전제조건
레퍼런스
Buçinca et al. (2021); Shneiderman (2020)
11 아바타 입동기 · 제스처 Embodiment effect
봇 기능
HeyGen Studio Avatar의 박대근 교수 본인 영상 기반 자연스러운 입동기와 제스처 생성
코드 위치
HeyGen 외부 서비스 + App.jsx startAvatar()
의도
Embodiment effect를 통한 사회적 실재감 강화
레퍼런스
Bickmore & Picard (2005); Mayer & DaPra (2012)
12 모드 전환 부드러움 (FTF/STS/TTT) Channel agency
봇 기능
동일 UI에서 좌하단 토글로 3 모달리티 즉시 전환 + 자원 자동 정리
코드 위치
App.jsx changeConversationMode()
의도
사용자에게 채널 선택권 부여 — SIP / MAIN model 적용
레퍼런스
Walther (1996)

Layer 4. 신뢰 메타 신호

6 / 18
봇 정책·관계·문화적 정합. 사용자가 의식적으로 인지하기 쉬워 설문 항목으로 가장 적합.
13 개인정보 동의 UI Transparency
봇 기능
필수(카카오 정보·개인정보)와 선택(마케팅) 동의를 분리하고 수집항목·이용목적·보유기간 노출
코드 위치
AuthModal.jsx consent section
의도
투명성(transparency) 신호
레퍼런스
Acquisti et al. (2015); Liao & Sundar (2022)
14 비로그인 둘러보기 옵션 Initial trust 진입로
봇 기능
"로그인 없이 둘러보기" 버튼으로 신뢰 형성 전 저-커밋 진입로 제공
코드 위치
AuthModal.jsx guestBtn
의도
초기 신뢰 형성(initial trust)의 핵심 설계 원칙
레퍼런스
McKnight et al. (2002)
15 한글 서수 인사말 Personalization
봇 기능
방문 횟수를 자연 한국어 서수("첫·두·열한·스물한번째")로 발화
코드 위치
App.jsx getKoreanVisitOrdinal(), getVisitGreeting()
의도
개인 인식(personalization)을 통한 사회적 단서
레퍼런스
Sundar & Marathe (2010)
16 방문 횟수 추적 Para-social continuity
봇 기능
MySQL users.visit_count 매 로그인마다 +1, 인사말에 반영
코드 위치
server/api.php handleKakaoLogin/handleEmailLogin / DB users 테이블
의도
일회성 거래가 아닌 지속적 관계로 프레이밍 — 비동기 para-social 메커니즘
레퍼런스
Horton & Wohl (1956); Lombard & Xu (2021)
17 TTS 발음 normalization 청각적 일관성
봇 기능
50+ 변환 규칙 — "AI"→"에이아이", "차의과학대학교"→"차 의과학 대학교", "GPT"→"지피티" 등
코드 위치
App.jsx normalizeTtsText()
의도
영문 약어와 합성어를 한국어 음운으로 자연스럽게 발음 — 청각적 일관성과 인식 부하 감소
레퍼런스
(DOI 검증된 직접 레퍼런스 추후 보강)
18 카카오 인앱 → 외부 브라우저 자동 전환 기술적 honesty · 접근권
봇 기능
카카오톡 인앱 브라우저 감지 시 Android Chrome intent / iOS kakaotalk://web/openExternal로 강제 전환
코드 위치
index.html 최상단 IIFE 스크립트
의도
마이크·WebRTC 권한 필요성을 솔직히 공개하고 사용자 접근권 보존
레퍼런스
Shneiderman (2020)

참고문헌 (DOI 검증)

아래 문헌은 모두 공식 발행처 DOI가 확인된 것입니다. 항목 #17(TTS normalization)의 직접 레퍼런스는 추후 보강이 필요합니다.

  1. Acquisti, A., Brandimarte, L., & Loewenstein, G. (2015). Privacy and human behavior in the age of information. Science, 347(6221), 509-514. https://doi.org/10.1126/science.aaa1465
  2. Bhattacherjee, A. (2002). Individual trust in online firms: Scale development and initial test. Journal of Management Information Systems, 19(1), 211-241. https://doi.org/10.1080/07421222.2002.11045715
  3. Bickmore, T. W., & Picard, R. W. (2005). Establishing and maintaining long-term human-computer relationships. ACM Transactions on Computer-Human Interaction, 12(2), 293-327. https://doi.org/10.1145/1067860.1067867
  4. Buçinca, Z., Malaya, M. B., & Gajos, K. Z. (2021). To trust or to think: Cognitive forcing functions can reduce overreliance on AI in AI-assisted decision-making. Proceedings of the ACM on Human-Computer Interaction, 5(CSCW1), Article 188. https://doi.org/10.1145/3449287
  5. Cowan, B. R., Pantidi, N., Coyle, D., Morrissey, K., Clarke, P., Al-Shehri, S., Earley, D., & Bandeira, N. (2017). "What can I help you with?": Infrequent users' experiences of intelligent personal assistants. Proceedings of MobileHCI '17. https://doi.org/10.1145/3098279.3098539
  6. Doney, P. M., & Cannon, J. P. (1997). An examination of the nature of trust in buyer-seller relationships. Journal of Marketing, 61(2), 35-51. https://doi.org/10.1177/002224299706100203
  7. Horton, D., & Wohl, R. R. (1956). Mass communication and para-social interaction. Psychiatry, 19(3), 215-229. https://doi.org/10.1080/00332747.1956.11023049
  8. Hovland, C. I., & Weiss, W. (1951). The influence of source credibility on communication effectiveness. Public Opinion Quarterly, 15(4), 635-650. https://doi.org/10.1086/266350
  9. Jacovi, A., Marasović, A., Miller, T., & Goldberg, Y. (2021). Formalizing trust in artificial intelligence: Prerequisites, causes and goals of human trust in AI. Proceedings of FAccT '21, 624-635. https://doi.org/10.1145/3442188.3445923
  10. Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., & Fung, P. (2023). Survey of hallucination in natural language generation. ACM Computing Surveys, 55(12), Article 248. https://doi.org/10.1145/3571730
  11. Lee, J. D., & See, K. A. (2004). Trust in automation: Designing for appropriate reliance. Human Factors, 46(1), 50-80. https://doi.org/10.1518/hfes.46.1.50.30392
  12. Liao, Q. V., & Sundar, S. S. (2022). Designing for responsible trust in AI systems: A communication perspective. Proceedings of FAccT '22, 1257-1268. https://doi.org/10.1145/3531146.3533182
  13. Lombard, M., & Xu, K. (2021). Social responses to media technologies in the 21st century: The Media Are Social Actors paradigm. Human-Machine Communication, 2, 29-55. https://doi.org/10.30658/hmc.2.2
  14. Mayer, R. E., & DaPra, C. S. (2012). An embodiment effect in computer-based learning with animated pedagogical agents. Journal of Experimental Psychology: Applied, 18(3), 239-252. https://doi.org/10.1037/a0028616
  15. McKnight, D. H., Choudhury, V., & Kacmar, C. (2002). Developing and validating trust measures for e-commerce: An integrative typology. Information Systems Research, 13(3), 334-359. https://doi.org/10.1287/isre.13.3.334.81
  16. Pataranutaporn, P., Danry, V., Leong, J., Punpongsanon, P., Novy, D., Maes, P., & Sra, M. (2021). AI-generated characters for supporting personalized learning and well-being. Nature Machine Intelligence, 3, 1013-1022. https://doi.org/10.1038/s42256-021-00417-9
  17. Pelau, C., Dabija, D.-C., & Ene, I. (2021). What makes an AI device human-like? The role of interaction quality, empathy and perceived psychological anthropomorphic characteristics in the acceptance of artificial intelligence in the service industry. Computers in Human Behavior, 122, 106855. https://doi.org/10.1016/j.chb.2021.106855
  18. Shneiderman, B. (2020). Human-centered artificial intelligence: Reliable, safe & trustworthy. International Journal of Human-Computer Interaction, 36(6), 495-504. https://doi.org/10.1080/10447318.2020.1741118
  19. Skantze, G. (2021). Turn-taking in conversational systems and human-robot interaction: A review. Computer Speech & Language, 67, 101178. https://doi.org/10.1016/j.csl.2020.101178
  20. Sundar, S. S., & Marathe, S. S. (2010). Personalization versus customization: The importance of agency, privacy, and power usage. Human Communication Research, 36(3), 298-322. https://doi.org/10.1111/j.1468-2958.2010.01377.x
  21. Walther, J. B. (1996). Computer-mediated communication: Impersonal, interpersonal, and hyperpersonal interaction. Communication Research, 23(1), 3-43. https://doi.org/10.1177/009365096023001001