Automated Matching of User Preferences and Product Components for Personalized Customization

There is a growing interest in “Kansei value,” which appeals to users’ senses and emotions, as an important factor in making a product unique. Therefore, many studies on impressions of “Kansei value” have been conducted. However, one of the problems with these studies is that they tend to focus on the product as a whole and lack detailed analyses of its components. The purpose of this study is to realize product customization according to user preferences by quantitatively evaluating the impressions that the product components give to people and clarifying the relation among them. This paper proposes a method for quantitatively evaluating the relations between product components and impression words. The method measures the strength of this relation by utilizing natural language processing techniques, specifically ConceptNet Numberbatch and Normalized Google Distance (NGD). In addition, the OpenAI API is used to decompose the product into its components, enabling a data-driven component segmentation without relying on human subjectivity. An evaluation experiment confirmed that both methods successfully matched impression words with product components. The matching method using ConceptNet Numberbatch clarified the relation between impressions and product components, which is difficult to capture by conventional methods, while also reflecting the technical significance of the components. Meanwhile, the NGD-based matching method exhibited a low probability of missing evaluation items and demonstrated a high level of correspondence.