Abstract
This study aimed to estimate psychological utility functions that convert product design into impressions using preference learning methodology, and to construct product shapes that match product concepts based on the estimated functions. Using Elliptic Fourier descriptors (EFDs), we converted the contours of 26 shampoo bottle shapes into Fourier coefficients and performed a Principal Component Analysis (PCA) on the coefficients to construct shape space. Twelve persons participated in five experimental sessions corresponding to five different product concepts. For each session, participants were presented with a pair of randomly generated images of shampoo bottles from the shape space. They were asked to choose the one that matched a given product concept for 100 trials. The bottle shapes conforming to the product concepts were synthesized based on the average utility functions estimated by using Gaussian process preference learning. The synthesized bottle shapes were assessed to determine if they conveyed the intended product concepts. The results suggested that our approach is an effective way to reflect the product concept in the shape design.
