文献类型：期刊文献

英文题名：Hybrid Kansei-Visual Interaction Framework for Human-Centered Product Form Design Based on SHAP-Kano and Transformer-BiLSTM

作者：Yang, Chaoxiang[1,2];Jin, Kaixiang[1];Ye, Junnan[1]

机构：[1]East China Univ Sci & Technol, Sch Art Design & Media, 130 Meilong Rd, Shanghai 200237, Peoples R China;[2]Art Dev Ctr Minist Culture & Tourism, Digital Art & Blockchain Lab, Beijing, Peoples R China

年份：2026

外文期刊名：INTERNATIONAL JOURNAL OF HUMAN-COMPUTER INTERACTION

收录：;EI(收录号：20262520954577);WOS:【SSCI(收录号:WOS:001819230000001)，SCI-EXPANDED(收录号:WOS:001819230000001)】；

基金：The authors wish to thank the Humanities and Social Science Fund of Ministry of Education of China and the Humanities and Social Science Youth Fund of Ministry of Education of China, CoDesign for financially supporting under Contract No. 25YJA760121 and No. 23YJC760144.

语种：英文

外文关键词：Kansei Engineering; visual sequence; eye tracking; online reviews; explainable machine learning

摘要：In the experience economy, user attention in product interaction has shifted from functional performance toward emotional and perceptual experience, making Kansei factors increasingly important in product form design (PFD). However, the identification and prioritization of customer requirements (CRs) often rely on subjective judgment, and form design decisions remain constrained by designers' intuition. This study proposes a hybrid human-centered Kansei Engineering (KE) framework that integrates explainable machine learning with visual interaction analysis. Online user reviews are utilized to objectively extract and evaluate affective requirements, while visual sequence features derived from eye-tracking experiments capture users' perceptual and attentional responses to product forms. Key CRs are identified using Latent Dirichlet Allocation, and sentiment information is analyzed through BERT and SnowNLP. An eXtreme Gradient Boosting (XGBoost) model combined with Shapley Additive Explanations (SHAP) is employed to perform Kano-based requirement classification and importance weighting. Subsequently, a Transformer-BiLSTM model maps Kansei factors to form features based on users' visual sequence data, enabling the prediction of form attribute combinations aligned with user affective expectations. A Bluetooth speaker design case study demonstrates the applicability of the proposed framework. Results indicate that integrating affective requirement modeling with visual attention analysis effectively supports data-driven and human-centered product form design, contributing to research on visual perception and interaction in human-computer interaction contexts.