Automatic Impression Indexing based on Evaluative Expression Dictionary from Review Data
Atsuhiro Yamada, Sho HASHIMOTO, Noriko Nagata · Transactions of Japan Society of Kansei Engineering · 2018
To meet user's affective needs, a number of efforts have been addressed to index kansei. However, the conventional method based on subjective evaluation experience take much time and effort to experiment and analyze. To solve this problem, we propose a method that automatically indexes impressions which is the first phase of attempting to index kansei based on evaluative expression dictionaries from the review data. This method consists of three steps. First, we collect evaluation words and categorize them as impression or emotion using word classes and the evaluative expression dictionary. Second, we extract the impression topics using the topic model, which uses only the evaluation words of impressions. Finally, we score each product for each impression topic, using the frequencies of evaluation words and term-scores. The results of an application of the method to the review texts of wristwatches and the subjective evaluation experiment show the validity of our method.