Extracting Kansei Evaluation Index Using Time Series Text Data: Examining Universality and Temporality

Abstract

In recent years, affective needs, such as usability and comfort, have attracted attention, along with conventional manufacturing needs, such as function, price and reliability. Therefore, when designing products, it is necessary to accurately and efficiently reflect user’s affective needs on product design. To that end, we clarify the user’s emotions and impressions, in terms of the ways that people feel about products. However, they are assumed to be constant regardless of time-series, impressions that are influenced by time-series and impressions that are used universally within a certain period are mixed. As the user’s emotions and impressions depend on time-series, it is necessary to deal with them separately, by time-series. In this study, based on time-series changes in the appearance frequency of evaluation words, we classified the Kansei evaluation index, according to whether it changes by time-series or not. In the proposed method, for each evaluation word, a state-space model is first used to extract the information of seasonal and trend variations. Second, by clustering this information, the evaluation terms are separated into several clusters. This method was applied to the fashion field, where time-series effects are believed to exist. The results confirmed that there were two patterns of seasonal variation and four patterns of trend variation in the impression of fashion in general, and eight universal impressions were extracted.