Structuring the Part-whole Relationship of Impression by Automatic Indexing of Kansei based on Review Texts

Abstract

In recent years, useful information for product design has been extracted from online reviews that describe evaluations or opinions on products. Since the product evaluation on these reviews often consists of partial impressions and the whole impression that synthesize them, the user's impression of the product has a part-whole relationship. If one or only a few parts relate to the whole impression, the part that determines the whole impression is clear. However, in the case of products consisting of many parts, since each impression of multiple parts affects the whole impression, the parts attributed to changing the whole impression are not clear, and this makes it difficult to achieve an arbitrary impression on the design. In this study, we propose a method of structuring the part-whole relationship of impressions based on reviews in order to acquire product knowledge that easily achievable on the product design. The method consists of 4 steps; (1) extracting the main whole impression in the target product domain; (2) extracting the main impression of each part in the target product domain; (3) analyzing the impact of the main impressions of each part on the main whole impressions; and (4) extracting elements related to the main impression of each part. Analysis of the actual data has shown that some main whole impressions of the product can be estimated from the partial impressions.