CLP at SemEval-2019 Task 3: Multi-Encoder in Hierarchical Attention Networks for Contextual Emotion Detection

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 164–168 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 164 CLP at SemEval-2019 Task 3: Multi-Encoder in Hierarchical Attention Networks for Contextual Emotion Detection Changjie Li Artificial Intelligence Group Sogou Beijing, China, 100000 lichangjie0308@gmail.com Yun Xing Artificial Intelligence Lab Lenovo Research, Lenovo Beijing, China, 100000 xingyun44@hotmail.com Abstract In this paper, we describe the participation of team ”CLP” in SemEval-2019 Task 3 “Con- textual Emotion Detection in Text” that aims to classify emotion of user utterance in tex- tual conversation. The submitted system is a deep learning architecture based on Hier- archical Attention Networks (HAN) and Em- bedding from Language Model (ELMo). The core of the architecture contains two represen- tation layers. The first one combines the out- puts of ELMo, hand-craft features and Bidi- rectional Long Short-Term Memory with At- tention (Bi-LSTM-Attention) to represent user utterance. The second layer use a Bi-LSTM- Attention encoder to represent the conversa- tion. Our system achieved F1 score of 0.7524 which outperformed the baseline model of the organizers by 0.1656. 1 Introduction Emotion detection has been widely researched in psychology, sociology and computer science. Be- ing able to recognize the emotion of text is of vital importance in the human-computer interac- tion (Cowie et al., 2001). However, detecting emotion in text is generally considered very chal- lenging in absence of facial expression or voice modulation. In domain of natural language pro- cessing, emotion detection is a task of associating words, phrases or documents with emotions us- ing psychological models (Duppada et al., 2018). Traditional rule-based approaches (Balahur et al., 2011; Chaumartin, 2007) and machine learning approaches (Alm et al., 2005; Balabantaray et al., 2012) rely on extracting word-level features to classify emotion. These methods suffer from low recall as many texts do not contain emotion words. To tackle the problem, recent deep learning ap- proaches (Mundra et al., 2017) take the advan- tage of Word2Vec representation (Mikolov et al., 2013a) to extract semantic features and achieve remarkable performance. However, limited re- searches have been done in classifying textual con- versation emotions, which is further compounded by difficulty in the context understanding. Task 3 “Contextual Emotion Detection in Text” in SemEval-2019 aims to find better solutions for those difficulties in contextual emotion de- tection (Chatterjee et al., 2019). The task con- siders textual emotion classification on four-point scale(Happy, Sad, Angry along with an Others category). It classifies emotion of user utterance along with 2 turns of context in conversation. This paper describes the components and results of our emotion recognition system. The proposed system is a deep learning model based on HAN, which combines multiple encoding methods in- cluding ELMo, hand-craft features and Bi-LSTM- Attention encoder. Our system yields a micro- averaged F1 score of 0.7524 on test-set of Task 3 of SemEval 2019. 2 System Description Figure 1 provides a overall architecture of our ap- proach, which consists of three components: (1) preprocessing, where we use a specially designed text …

Content truncated to 500 words