Extensive research has been conducted to develop emotion classification models as a way to effectively detect and analyze emotions; however, there is still room for improvement because of (1) reliability issues of emotion labels, (2) small amount of reliable data, and (3) lack of model application. This paper presents emotion research that addresses these issues. We used a large-scale, emotion-labeled text dataset (924,827 online posts) directly specified by the authors and evaluated its validity through comparisons with other representative emotion datasets. The emotion classification model yielded performance up to 81% accuracy. We applied our model to two popular social networking sites, Reddit and Yelp, and evaluated feasibility and challenges to be considered in the application of emotion modeling. Especially, our study results highlight the ambiguity of the love emotion, and we discuss how to deal with it from theoretical perspectives. Finally, we present a case study of using emotion models to understand consumer needs.
