Construction of prediction models for perceptual imagery of product shape by GABPNN and SOA-BPNN

To improve the efficiency and scope of vocabulary and sample collection in traditional Kansei engineering (KE), and to address the issue of poor accuracy in predicting product form imagery, this study introduces a novel research framework that incorporates elements of artificial intelligence, natural language processing, machine learning, and optimization algorithms. Initially, product samples and Kansei vocabulary were collected from e-commerce platforms using Python web crawlers and Word2vec, with representative samples and vocabulary selected. Following this, Kansei prediction models were developed using Back Propagation Neural Network (BPNN), a nonlinear machine learning method, and enhanced with the application of the Seagull Optimization Algorithm (SOA) and Genetic Algorithm (GA) to improve their performance. Lastly, an error comparison of the prediction models was conducted to determine the optimal model for guiding product form design, with hair dryer products used as the case study for validation. The results indicate that the SOA-BPNN model had the smallest prediction error, demonstrating that employing SOA to enhance BPNN's performance in predicting users’ emotional responses to product form is reliable. The SOA-BPNN model also showed greater robustness, making it suitable for predicting form imagery in product design. Moreover, this research framework provides valuable insights for related studies in KE.