In order to identify the core demand elements of smart desk users, the mapping relationship between product design elements and user demand perception was explored,which provides a scientific basis for the design optimization of smart desks. A total of 30 015 online reviews of smart desks were collected from Taobao and Jingdong platforms. After data cleaning and word segmentation, the latent dirichlet allocation (LDA) model was applied to cluster user needs into four major categories and thirty-two subcategories. These demand categories were treated as exogenous latent variables and the subcategories as observed variables. Questionnaire data were collected using a five-point Likert scale, and a hybrid model combining structural equation modeling and artificial neural networks (SEM-ANN) was established. The SEM analysis indicated that the positive effect of “experience of use” on “satisfaction” was the most significant. The R² values of the SEM-ANN model increased by about 0.15~0.18 compared with the SEM model, while the RMSE(root mean square error)decreased significantly, showing enhanced nonlinear fitting ability. The comprehensive weight analysis identified odor and environmental friendliness, surface touch, lift safety, and self-assembly installation as the key design optimization points. By integrating online review mining with SEM-ANN hybrid modeling, the mapping between micro design elements and macro user perception of smart lifting desks was revealed. The proposed method provides a prioritized improvement list for design optimization and can be extended to user requirement analysis of other smart hardware and complex consumer products.