A Text Mining Approach for Automatic Modeling of Kansei Evaluation from Review Texts

In the field of product design, it is important to meet the specific user’s affective needs in addition to function and price. For that purpose, a Kansei evaluation model expressing the relationships between a low-level impression related to physical characteristics and a high-level impression (affective needs) is being constructed in the field of Kansei engineering. However, the conventional modeling method involves a lot of time and effort, because it depends on experiments involving humans and on subsequent analyses. To solve this problem, we propose a method that automatically constructs a Kansei evaluation model for product design using review texts on the Web. The method consists of three steps. First, we collect and select evaluation words with word classes and Japanese dictionaries of evaluation expressions. Second, we estimate the impression axes particular to a domain with the topic model that uses only the evaluation words. Finally, we score each product for each evaluation axis, using the frequencies of the appearance of evaluation words and term-scores. The results of the application of the method to reviews of wristwatches and the subjective evaluation experiment with topics for visual impressions show the utility and validity of the method.