KANO-QSL 통합모형기반 생성형 AI 안내 시스템의 체험 품질 및 이용자 만족도 연구 = A Study on Experience Quality and User Satisfaction of Generative AI Guide Systems Based on the KANO QSL － Integrated Model

다국어 초록 (Multilingual Abstract)

Generative AI guide systems are increasingly being introduced into public cultural spaces such as libraries, museums, and exhibition venues. Unlike conventional guide systems that rely on pre-defined menus or fixed retrieval structures, generative AI systems can provide real-time responses, contextual explanations, and personalized recommendations through conversational interaction. However, existing evaluations of these systems have tended to focus on technical performance, such as accuracy, response speed, and recommendation efficiency. Less attention has been paid to how users experience generative AI guide systems in actual cultural spaces, how different service attributes contribute to satisfaction or dissatisfaction, and how perceived quality is translated into behavioral intention. Addressing this gap, this study examines the experience quality and user satisfaction of the generative AI guide system “TU YUEYUE（图悦阅）” installed in the Beijing City Library, using an integrated KANO－QSL model.
The purpose of this study is threefold. First, it identifies and classifies the major service attributes of the generative AI guide system through the KANO model, with particular attention to the asymmetric relationship between satisfaction and dissatisfaction. Second, it examines the structural relationships among functional quality, experiential quality, cognitive satisfaction, emotional-experiential satisfaction, continuance intention, and recommendation intention within a QSL framework. Third, it integrates the attribute-level results of the KANO analysis with the structural path results of the QSL analysis in order to derive implications for the design and operation of generative AI guide systems in digital cultural spaces. Data were collected from users who had directly experienced the “TU YUEYUE” system. A total of 258 valid responses were used for the final analysis. The data were analyzed using descriptive statistics, reliability analysis, correlation analysis, confirmatory factor analysis, structural equation modeling, and bootstrap-based parallel mediation analysis. KANO classification, Better-Worse coefficients, and PCSI analysis were also applied to identify user requirements and improvement priorities.
The KANO analysis showed that the service attributes of the generative AI guide system did not operate in a uniform or linear manner. Attributes related to information accuracy, response speed, and ease of operation were closely associated with the functional stability of the system and played an important role in reducing dissatisfaction. By contrast, attributes such as natural conversation, personalized recommendation, and interesting interaction were more closely related to experiential engagement and differentiated user value. This indicates that users evaluate generative AI guide systems not only as information tools but also as interactive experience interfaces within the spatial and service context of a public library.
The QSL path analysis further revealed that both functional quality and experiential quality had significant positive effects on cognitive satisfaction and emotional-experiential satisfaction. However, the expected distinction between the two quality paths was not clearly supported. Functional quality did not have a stronger effect on cognitive satisfaction than on emotional-experiential satisfaction, and experiential quality did not have a clearly stronger effect on emotional-experiential satisfaction than on cognitive satisfaction. This result suggests that, in the context of generative AI guide systems, users tend to perceive functional reliability and experiential interaction as closely connected. Accurate and smooth responses may provide not only cognitive assurance but also emotional reassurance, while natural and personalized interaction may reinforce users’ cognitive evaluation of the system.
Both cognitive satisfaction and emotional-experiential satisfaction had significant positive effects on continuance intention and recommendation intention. The bootstrap-based mediation analysis confirmed that the two satisfaction dimensions mediated the relationship between perceived quality and behavioral intention. The overall parallel mediation effect was therefore supported. However, the relative dominance of specific mediation paths was limited. Functional quality was not mediated mainly through cognitive satisfaction alone, and experiential quality was only conditionally associated with emotional-experiential satisfaction as the dominant pathway. These findings suggest a dual but interrelated satisfaction structure in which behavioral intention is formed through the combined operation of cognitive and emotional-experiential evaluation.
This study contributes to research on digital cultural spaces and AI-based service experience by combining the KANO model and the QSL framework in a single empirical design. The KANO model clarifies which service attributes should be treated as basic conditions, performance factors, or attractive elements, while the QSL framework explains how perceived quality leads to satisfaction and behavioral intention. By integrating the two approaches, this study connects attribute-level improvement priorities with the broader structure of user response. In practical terms, the findings suggest that the UI/UX improvement of generative AI guide systems should proceed in stages. Basic usability, including clear menu structure and operational clarity, should first be secured. Performance-related qualities such as information accuracy, response speed, natural dialogue, and personalized recommendation should then be strengthened through visible information sources, continuous conversational flow, and clear recommendation reasons. Attractive experiential elements, such as novelty and enjoyable interaction, should be applied carefully so that they enrich the user experience without disrupting the quiet and information-oriented context of a public library. The study is limited to a single case and cross-sectional survey data. Future research should compare different cultural spaces and AI guide systems and further examine longitudinal changes in user perception as generative AI technologies continue to evolve.

목차 (Table of Contents)

• 1. 서론 1
• 1.1 연구 배경 및 목적 1
• 1.2. 연구 범위 및 방법 4
• 1.3. 선행연구 검토 8
• 1.4. 연구의 흐름도 14
• 2. 디지털 문화공간 안내 시스템 분석 프레임워크 15
• 2.1. 디지털 문화공간의 개념과 체험 구조 15
• 2.1.1. 디지털 문화공간의 개념 15
• 2.1.2. 디지털 문화공간의 유형 16
• 2.1.3. 디지털 문화공간의 핵심 요소 19
• 2.2. 생성형 AI 안내 시스템의 구조와 상호작용 특성 20
• 2.2.1. AI 기술의 진화와 생성형 AI의 등장 21
• 2.2.2. 생성형 AI 안내 시스템의 구조 21
• 2.2.3. 연구대상 시스템 ‘도열열(图悦阅)’의 서비스 구조와 사용 장면 25
• 2.2.4. 생성형 AI의 문화공간에서의 역할 33
• 2.2.5. 생성형 AI 안내 시스템의 문화공간적 특징 34
• 2.3. 생성형 AI 안내 시스템의 체험 품질 구성 37
• 2.3.1. 서비스 품질의 개념과 기능적 품질 37
• 2.3.2. 체험 품질의 개념과 경험적 품질 39
• 2.3.3. 기능적 품질과 경험적 품질의 통합 필요성 41
• 2.4. 사용자 만족도와 품질–만족–행동의도(QSL) 경로 43
• 2.4.1. 인지적 만족 43
• 2.4.2. 감정적·체험적 만족 44
• 2.4.3. 만족 차원과 행동의도 44
• 2.5. 사용자 요구 분석을 위한 KANO 모델 45
• 2.5.1. KANO 모델의 품질 속성 분류 45
• 2.5.2. Timko의 고객만족계수 및 PCSI 지수 48
• 2.5.3. 생성형 AI 안내 시스템 속성과 KANO 분류 50
• 2.6. KANO–QSL 통합모형의 이론적 구조와 분석 프레임워크 51
• 2.6.1. 속성 진단과 경로 검증의 결합 51
• 2.6.2. 통합 분석 프레임워크 51
• 3. 연구 설계 및 방법 54
• 3.1. 연구 모형 및 가설 설정 54
• 3.1.1. 연구 모형의 구성 54
• 3.1.2. 연구문제 및 가설 설정 55
• 3.2. 변수의 조작적 정의 및 측정 도구 개발 56
• 3.2.1. 체험 품질(Q), 만족(S) 및 행동의도(L)의 조작적 정의 및 측정 57
• 3.2.2. KANO 속성 측정 59
• 3.2.3. 인구통계학적 및 이용 특성 변수의 측정 61
• 3.2.4. 설문지 구성 및 측정 도구 검토 61
• 3.3. 조사 대상 및 표본 설계 62
• 3.3.1. 조사 대상 및 표집 방법 62
• 3.3.2. 설문 절차 및 자료 수집 63
• 3.3.3. 표본 선정 기준 및 자료 정제 64
• 3.3.4. 표본 규모 및 대표성 64
• 3.4. 자료 분석 방법 64
• 4. 실증 분석 및 결과 67
• 4.1. 표본의 일반적 특성 67
• 4.2. KANO 모델 기반 사용자 요구 분석 71
• 4.2.1. KANO 분류 결과 71
• 4.2.2. Better-Worse 계수 분석 75
• 4.2.3. PCSI 지수 분석 79
• 4.3. 품질–만족–행동의도(QSL) 경로 분석 82
• 4.3.1. 측정도구의 신뢰도 검증 82
• 4.3.2. 상관관계 분석 83
• 4.3.3. 확인적 요인분석 CFA 85
• 4.3.4. SEM 경로분석 89
• 4.3.5. Bootstrap 병렬매개효과 분석 93
• 4.4. 통합 분석 결과 96
• 4.5. 가설 검증 결과 요약 99
• 5. 결론 및 제언 101
• 5.1. 연구 결과 101
• 5.2. 결론 및 시사점 104
• 5.2.1. 학문적 시사점 104
• 5.2.2. 실무적 및 디자인적 시사점: ‘도열열(图悦阅)’ UI/UX 개선 방향 106
• 5.3. 연구의 한계 및 향후 연구 방향 109
• 참고문헌 114
• 부 록 124
• Abstract 132