Emotion Analysis with a Writer’s Multiple Posts

Abstract

Emotion analysis is a natural language processing task that estimates a writer's sentiment polarity or emotional intensity from text. To personalize emotion analysis to the writer, previous studies have taken into account the bias in the distribution of emotion labels and the personality of the writer. However, these previous studies are costly for annotations such as emotion labeling and personality testing by the target writers. To address this issue, we propose a method to personalize emotion analysis without additional annotation. To capture the characteristics of each writer, our method inputs the text to be analyzed along with multiple other texts by that writer. Experimental results on emotional intensity estimation in Japanese and sentiment polarity classification in both Japanese and English revealed the effectiveness of the proposed method for personalized emotion analysis.