Evaluating and enhancing creativity in the design of museum cultural and creative products (MCCPs)

: measurement, mechanisms, prediction, and human-AI collaboration

Abstract

Museum cultural and creative products (MCCPs) have become vital carriers of cultural dissemination, aesthetic innovation, and economic sustainability. Yet, the concept of “creativity” in MCCP design remains difficult to define, measure, and operationalize. Existing models emphasize novelty or functionality but often neglect the affective, aesthetic, and cultural interpretations central to consumer responses. This thesis addresses these gaps by asking: (1) How can creativity in MCCPs be effectively defined and measured from the consumer perspective? (2) What perceptual and cognitive mechanisms shape consumer evaluations of creativity? (3) How accurately can creativity be predicted using machine learning trained on structured evaluation data? (4) What human factors influence the adoption and effectiveness of AI-generated feedback in real-world design?
To answer these questions, the thesis proposes a systematic, multi-method program organized into seven interlinked empirical studies. Drawing on consumer surveys, review analysis, predictive modeling, and A/B testing with designers, the research develops and validates a domain-specific framework for creativity evaluation. Early studies identify five empirically supported dimensions—Novelty, Usefulness, Affect, Aesthetics, and Cultural Values—captured in the Museum Product Creativity Measurement (MPCM) model. Robustness tests confirm the dimensional structure, while further analysis uncovers a dual-pathway mechanism: perceptual cues (Aesthetics, Affect, Cultural Values) exert more immediate influence, whereas cognitive cues (Novelty, Usefulness) are moderated by user background and context. Building on this foundation, machine learning models—including Random Forest, GBDT, LightGBM, XGBoost, and SVR—are applied to predict creativity scores. Results show that Random Forest achieves reliable prediction even with relatively small training samples, highlighting scalability. Finally, the MuseCrea system, an AI-powered feedback tool, is tested through three rounds of A/B experiments involving 13 designers and 1,478 consumer participants. Findings reveal that system effectiveness depends not only on algorithmic accuracy but also on designer motivation, cognitive adoption, feedback uptake, and the perceptual salience of design revisions.
The research contributes theoretically by constructing a domain-specific, multidimensional model of MCCP creativity, clarifying the dual cognitive–perceptual pathways of consumer evaluation, and demonstrating the predictive validity of machine learning grounded in consumer data. Methodologically, it integrates large language model–assisted sentiment analysis, structural equation modeling, and predictive algorithms, advancing a scalable approach to creativity assessment. Empirically, it provides robust evidence from large-scale datasets and experimental testing. Practically, the thesis operationalizes these insights in MuseCrea, offering museums and designers an interpretable, AI-powered feedback system to address creativity stagnation and product homogenization in the cultural and creative industries. Future research may refine prediction under minimal data conditions, extend AI-designer collaboration models to real-world contexts, and explore LLM-based tools capable of evaluating creativity directly from product images, offering scalable alternatives to traditional survey-based methods.

Date of Award	15 Mar 2026
Original language	English
Awarding Institution	University of Nottingham
Supervisor	Bingjian Liu (Supervisor), Jing Xie (Supervisor), Shijian Luo (Supervisor) & Cheng Yao (Supervisor)