Design-Oriented User Experience Analysis of Consumer AI Glasses Using Large-Scale Online Reviews: Large Language Model-Assisted Aspect-Based Sentiment Analysis and Explainable Machine Learning

The functions and application scenarios of AI glasses continue to expand; however, research on design optimization informed by user needs and usage experiences remains limited. This study develops a natural language processing framework for design optimization that integrates BERTopic, large language model–based EIP-CABSA, and explainable machine learning. A total of 17,116 user reviews of AI glasses were collected from JD. A large language model was then used to construct a structured analytical matrix, while platform ratings were employed as an observed measure of overall user satisfaction. The results indicate that CatBoost achieved the best predictive performance among the candidate models. AI Interaction and Intelligence, together with Imaging and Audio Perception, exhibited high global importance, whereas After-sales and Support were characterized by low coverage but high importance. Connection and Ecosystem, together with Workmanship and Durability, showed high conditional importance, while privacy-related issues were mentioned infrequently but were associated with a concentration of negative evaluations. Based on these predictive associations, this study outlines a tiered set of design priorities for subsequent evaluation. It provides an interpretable quantitative approach for identifying nonlinear associations between design aspects and overall user satisfaction, thereby supporting the design optimization of consumer-grade AI glasses.