Emotion Distribution Learning from Texts

Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 638–647, Austin, Texas, November 1-5, 2016. c⃝2016 Association for Computational Linguistics Emotion Distribution Learning from Texts Deyu Zhou, Xuan Zhang, Yin Zhou, Quan Zhao, Xin Geng∗ MOE Key Laboratory of Computer Network and Information Integration School of Computer Science and Engineering Southeast University, Nanjing, China {d.zhou, zhouying1, xuanzhang, zhaoquan, xgeng}@seu.edu.cn Abstract The advent of social media and its prosperity enable users to share their opinions and views. Understanding users’ emotional states might provide the potential to create new business opportunities. Automatically identifying user- s’ emotional states from their texts and clas- sifying emotions into finite categories such as joy, anger, disgust, etc., can be consid- ered as a text classification problem. How- ever, it introduces a challenging learning sce- nario where multiple emotions with differen- t intensities are often found in a single sen- tence. Moreover, some emotions co-occur more often while other emotions rarely co- exist. In this paper, we propose a novel ap- proach based on emotion distribution learning in order to address the aforementioned issues. The key idea is to learn a mapping function from sentences to their emotion distributions describing multiple emotions and their respec- tive intensities. Moreover, the relations of e- motions are captured based on the Plutchik’s wheel of emotions and are subsequently in- corporated into the learning algorithm in order to improve the accuracy of emotion detection. Experimental results show that the proposed approach can effectively deal with the emo- tion distribution detection problem and perfor- m remarkably better than both the state-of-the- art emotion detection method and multi-label learning methods. 1 Introduction The advent of social media and its prosperity enable the creation of massive online user-generated con- ∗Corresponding author Sentence Trains crash near Thai resort town Emotions anger disgust fear joy sadness surprise 2 0 62 0 90 10 Table 1: An example of a sentence containing emotions select- ed from SemEval 2007 Task#14, Affective Text, where each of the six emotions are indicated using a score of [0, 100]. tent including opinions and product reviews. Ana- lyzing such user-generated content allows the detec- tion of users’ emotional states, which might be po- tentially useful for downstream applications such as brand watching, product recommendation, and de- tection of health-related issues, etc. Based on the way emotions are represented, computational mod- els for emotion analysis can be categorized into di- mensional models and categorical models (Calvo and D’Mello, 2010). Dimensional approaches (Rus- sell, 2003) emphasize the fundamental dimension- s of valence and arousal in understanding emotion- al experience, which have long been studied by e- motion theorists. Categorical models (Gupta et al., 2013) involve the use of a categorical representa- tion, in which emotions are represented by a num- ber of labels. For example, Ekman’s basic emotion set (Ekman, 1992) consists of anger, disgust, fear, happiness, sadness and surprise. An example of a sentence and the annotated emotions can be found in Table 1. Considering each basic emotion as …

Content truncated to 500 words