CAiRE_HKUST at SemEval-2019 Task 3: Hierarchical Attention for Dialogue Emotion Classification

CAiRE HKUST at SemEval-2019 Task 3: Hierarchical Attention for Dialogue Emotion Classification Genta Indra Winata*, Andrea Madotto*, Zhaojiang Lin, Jamin Shin, Yan Xu, Peng Xu, Pascale Fung Center for Artificial Intelligence Research (CAiRE) Department of Electronic and Computer Engineering The Hong Kong University of Science and Technology, Clear Water Bay, Hong Kong {giwinata,amadotto,zlinao}@connect.ust.hk, {jmshinaa,yxucb,pxuab}@connect.ust.hk,pascale@ece.ust.hk Abstract Detecting emotion from dialogue is a chal- lenge that has not yet been extensively surveyed. One could consider the emo- tion of each dialogue turn to be indepen- dent, but in this paper, we introduce a hi- erarchical approach to classify emotion, hypothesizing that the current emotional state depends on previous latent emotions. We benchmark several feature-based clas- sifiers using pre-trained word and emo- tion embeddings, state-of-the-art end-to- end neural network models, and Gaussian processes for automatic hyper-parameter search. In our experiments, hierarchi- cal architectures consistently give signif- icant improvements, and our best model achieves a 76.77% F1-score on the test set. 1 Introduction Customer service can be challenging for both the givers and receivers of services, leading to emo- tions on both sides. Even human service-people who are trained to deal with such situations strug- gle to do so, partly because of their own emo- tions. Neither do automated systems succeed in such scenarios. What if we could teach machines how to react under these emotionally stressful sit- uations of dealing with angry customers? This paper represents work on the SemEval 2019 shared task (Chatterjee et al., 2019b), which aims to bring more research on teaching machines to be empathetic, specifically by contextual emo- tion detection in text. Given a textual dialogue with two turns of context, the system has to clas- sify the emotion of the next utterance into one of the following emotion classes: Happy, Sad, Angry, or Others. The training dataset contains *Equal contribution. 15K records for emotion classes, and contains 15K records not belonging to any of the aforemen- tioned emotion classes. The most naive first step would be to recognize emotion from a given flattened sequence, which has been researched extensively despite the very abstract nature of emotion (Socher et al., 2013; Felbo et al., 2017a; McCann et al., 2017; Xu et al., 2018; Chatterjee et al., 2019a). However, these flat models do not work very well on dialogue data as we have to merely concatenate the turns and flat- ten the hierarchical information. Not only does the sequence get too long, but the hierarchy be- tween sentences will also be destroyed (Hsu and Ku, 2018; Kim et al., 2018). We believe that the natural flow of emotion exists in dialogue, and us- ing such hierarchical information will allow us to predict the last utterance’s emotion better. Naturally, the next step is to be able to detect emotion with a hierarchical structure. To the best of our knowledge, this task of extracting emotional knowledge in a hierarchical setting has not yet been extensively explored in the literature. There- fore, in this paper, we investigate this problem in depth with …

Content truncated to 500 words