Estimating Emotion Intensities in Japanese Tweets Using Emotion Intensity Lexicon

Abstract

This paper presents a method for estimating emotion intensities in Japanese tweets using emotion intensity lexicon. We intend to utilize this method to create dialogue systems that can regulate user’s emotion. Such dialogue systems can be expected to be useful for maintaining the good mental condition and interpersonal relationships of the users. We first created a crowdsourced emotion intensity lexicon for eight categories of emotion (joy, sadness, anger, fear, trust, disgust, anticipation, surprise), which captures word–emotion intensities using best-worst scaling. To the best of our knowledge, there is no previous works to construct emotion intensity lexicon in Japanese. Then we carried out an experiment to evaluate the effect of the emotion intensity lexicon in the estimation of emotion intensities in Japanese tweets. The experimental results show that the emotion intensity lexicon can improve the accuracy of estimating emotion intensities in tweets.