Analysis of modifier structure for emotion expressions

Abstract

Dimensional emotion representation such as valence and arousal (VA) space has been an emerging way to represent emotions. In this representation, emotion words can be projected to the VA space according to their valence and arousal values. Sentence and document-level emotions can then be projected based on the emotion words within them. However, emotion expressions in sentences and documents usually contain various modifier structure such as negation (e.g., not happy), degree (very happy) and emotion compounds. Such modifier structure can provide more precise information for measuring VA values in both sentence and document-levels. In this study, we analyze various types of modifier structure for emotion expressions. In addition, we also investigate the effect of different types of modifier structure on measuring VA values for emotion expressions.