A Deep Learning-Based Approach to Facilitate Cross-cultural Kansei Design

With the development of economic globalization, more and more product designers are faced with the need of designing for customers from different countries. However, it is a challenge for designers to efficiently develop the desired mental images of certain products for target users with different cultural backgrounds. Our research proposed a deep learning-based system to facilitate designers to gain better awareness of the cross-culture differences between different target customers. We trained a kawaii classification neural network model with the data of 1414 cosmetic packaging images annotated by 12 Japanese females separately. As a follow-up investigation, we conducted neuron analysis to compare the features of kawaii packages perceived by Japanese participants with the results from a prior study conducted with Chinese participants. The result shows that Japanese females tended to see more girlish and exquisite design features as kawaii while Chinese females perceived more childish and round elements as kawaii. A reverse experiment further verified the effectiveness of adding these different design features to enhance Chinese or Japanese females’ perception of kawaii. We also noticed that it’s hard to obtain the cross-cultural differences in customers’ perception by extracting image parameters with a set of predefined visual features as such perception differences could be subconscious. Our deep learning-based Kansei design facilitation provides a feasible solution to customized design for target customers with different cultural backgrounds.