EmoNet: Fine-Grained Emotion Detection with Gated Recurrent Neural Networks

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 718–728 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1067 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 718–728 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1067 EmoNet: Fine-Grained Emotion Detection with Gated Recurrent Neural Networks Muhammad Abdul-Mageed School of Library, Archival & Information Studies University of British Columbia muhammad.mageed@ubc.ca Lyle Ungar Computer and Information Science University of Pennsylvania ungar@cis.upenn.edu Abstract Accurate detection of emotion from natu- ral language has applications ranging from building emotional chatbots to better un- derstanding individuals and their lives. However, progress on emotion detection has been hampered by the absence of large labeled datasets. In this work, we build a very large dataset for fine-grained emo- tions and develop deep learning models on it. We achieve a new state-of-the-art on 24 fine-grained types of emotions (with an average accuracy of 87.58%). We also extend the task beyond emotion types to model Robert Plutchik’s 8 primary emo- tion dimensions, acquiring a superior ac- curacy of 95.68%. 1 Introduction According to the Oxford English Dictionary, emo- tion is defined as “[a] strong feeling deriving from one’s circumstances, mood, or relationships with others.” 1 This “standard” definition identifies emotions as constructs involving something innate that is often invoked in social interactions and that aids in communicating with others(Hwang and Matsumoto, 2016). It is no exaggeration that hu- mans are emotional beings: Emotions are an in- tegral part of human life, and affect our decision making as well as our mental and physical health. As such, developing emotion detection models is important; they have a wide array of applications, ranging from building nuanced virtual assistants that cater for the emotions of their users to de- tecting the emotions of social media users in order to understand their mental and/or physical health. 1https://en.oxforddictionaries.com/ definition/emotion. However, emotion detection has remained a chal- lenging task, partly due to the limited availabil- ity of labeled data and partly due the controversial nature of what emotions themselves are (Aaron C. Weidman and Tracy, 2017). Recent advances in machine learning for natu- ral language processing (NLP) suggest that, given enough labeled data, there should be an oppor- tunity to build better emotion detection models. Manual labeling of data, however, is costly and so it is desirable to develop labeled emotion data without annotators. While the proliferation of social media has made it possible for us to ac- quire large datasets with implicit labels in the form of hashtags (Mohammad and Kiritchenko, 2015), such labels are noisy and reliable. In this work, we seek to enable deep learning by creating a large dataset of fine-grained emo- tions using Twitter data. More specifically, we harness cues in Twitter data in the form of emo- tion hashtags as a way to build a labeled emotion dataset that we then exploit using distant supervi- sion (Mintz et al., 2009) (the …

Content truncated to 500 words