Design of Impression Scales for Assessing Impressions of News Articles

This paper focuses on the impressions that people get from reading articles in newspapers. We have already proposed web application systems that extract and use several types of impressions from news articles. However, the types of impressions extracted and used in these systems were intuitively defined by us on the basis of a basic emotion model, which the well-known psychologist Robert Plutchik proposed to represent human emotions. That is, the characteristics of news articles that result in different impressions have not been taken into consideration in much detail. Therefore, we have tried to design one or more impression scales suitable for assessing impressions generated by news articles. First, we conducted nine experiments in each of which 100 people read ten news articles and indicated their impressions on 42 five-point scales, where 42 impression-related words such as “happy” and “strained” were assigned for the 42 scales. Consequently, we obtained impression-estimation data for the 42 impression-related words. Next, we applied factor and cluster analysis to these impression-estimation data, and analyzed similarities among the impression-related words in terms of their scores. In our results, the words that convey similar impressions are classified into a single group and the words that convey opposite impressions are classified into different groups of words. Finally, we designed six impression scales suitable for assessing impressions generated by news articles on the basis of these results, each of which consisted of two contrasting groups of impression-related words.