Emotion cause extraction, a challenging task with corpus construction

In this paper, we present a new challenging task for emotion analysis called emotion cause extraction. In this task, we do not need to identify the emotion category or emotion component of text. We focus on the emotion cause, a.k.a the reason or stimulant of an emotion. Since there is no open dataset available, the lack of annotated resources has limited the research in this area. Thus, we first built an annotated dataset for this task using SINA city news which follows the scheme of W3C Emotion Markup Language. We then present an emotion cause detection method using event extraction where a one-hot representation method is using to represent events in text. Because traditional event representation method does not consider the emotion category caused by the event, we modified the definition of event with a more reasonable improvement. Even with a limited training set, we can still extract sufficient features for analysis. Evaluations show that our approach achieves 7.68% higher F-measure than other reported methods. The contributions of our work include both resources and algorithm development.