Due to the inherent subjectivity of Kansei perception, aligning the front-end styling of new energy vehicles (NEVs) with users’ emotional preferences remains a complex challenge. This study proposes a data-driven framework integrating semantic mining and deep learning to quantify and optimize such emotional responses. A Latent Dirichlet Allocation (LDA) model was employed to extract four core emotional dimensions—fashion, power, technology, and sportiness—from a corpus of user-generated content (UGC). To establish a mapping between abstract emotions and concrete morphological features, rough set theory (RST) was applied for dimensionality reduction, retaining only the most influential design attributes. In addition, an attention-enhanced long short-term memory (LSTM) network optimized via a genetic algorithm (GA) was constructed to predict emotional evaluations. This hybrid model enables targeted design configuration generation for NEV front-end styling based on specific emotional indicators. The results demonstrate that the proposed approach effectively bridges the gap between qualitative user imagery and quantitative design features, providing the automotive industry with a robust emotion-oriented design support tool.
Yu et al. (Thu,) studied this question.