Tensor Self-Organizing Map for Kansei Analysis

In Kansei analysis, impressions of various objects are commonly measured using evaluation words. When using this approach, it is necessary to examine all combinations of three elements: subjects, objects, and evaluation words. However, the exhaustive analysis required is not an easy task because of the enormous number of combinations. Additionally, if it is necessary to reveal the relationship between the impressions and physical features of objects such as colors or shapes, the number of combinations increases enormously and the task becomes unrealistic. In this paper, we introduce a method called the tensor self-organizing map (TSOM) that visualizes the relationships between the elements. We applied the TSOM to Kansei analysis of landscape images and studied how the impressions were dependent on the subjects. We also investigated the relationships between these subject-dependent impressions and the physical features. Through these experiments, we demonstrate that the TSOM can be a useful tool for Kansei analysis.