Exploiting Unsupervised Data for Emotion Recognition in Conversations

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 4839–4846 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 4839 Exploiting Unsupervised Data for Emotion Recognition in Conversations Wenxiang Jiao† Michael R. Lyu† Irwin King† † Department of Computer Science and Engineering The Chinese University of Hong Kong, HKSAR, China † {wxjiao,lyu,king}@cse.cuhk.edu.hk Abstract Emotion Recognition in Conversations (ERC) aims to predict the emotional state of speak- ers in conversations, which is essentially a text classification task. Unlike the sentence-level text classification problem, the available super- vised data for the ERC task is limited, which potentially prevents the models from playing their maximum effect. In this paper, we pro- pose a novel approach to leverage unsuper- vised conversation data, which is more accessi- ble. Specifically, we propose the Conversation Completion (ConvCom) task, which attempts to select the correct answer from candidate an- swers to fill a masked utterance in a conver- sation. Then, we Pre-train a basic COntext- Dependent Encoder (PRE-CODE) on the Con- vCom task. Finally, we fine-tune the PRE- CODE on the datasets of ERC. Experimen- tal results demonstrate that pre-training on un- supervised data achieves significant improve- ment of performance on the ERC datasets, par- ticularly on the minority emotion classes.1 1 Introduction Emotion recognition in conversations (ERC) has garnered attention recently (Poria et al., 2019), due to its potential in developing practical chatting ma- chines (Zhou et al., 2018a). Unlike traditional text classification that handles context-free sentences, ERC aims to predict the emotional state of each utterance in a conversation (Figure 1). The inherent hierarchical structure of a conversation, i.e., words- to-utterance and utterances-to-conversation, deter- mines that the ERC task should be better addressed by context-dependent models (Poria et al., 2017; Hazarika et al., 2018b; Jiao et al., 2019, 2020). Despite the remarkable success, context- dependent models suffer from the data scarcity 1The source code is available at https://github. com/wxjiao/Pre-CODE You sprayed my front twice! You never turned? No! I barely even got to three Mississippi. Mississippi? I said count to five. [Angry] [Surprised] [Angry] [Neutral] Figure 1: A conversation example with emotion labels. issue. In the ERC task, annotators are required to recognize either obvious or subtle difference between emotions, and tag the instance with a spe- cific emotion label, such that supervised data with human annotations are very costly to collect. In addition, existing datasets for ERC (Busso et al., 2008; Hsu and Ku, 2018; Zahiri and Choi, 2018; Zadeh et al., 2018) contain inadequate conversa- tions, which prevent the context-dependent models from playing their maximum effect. In this paper, we aim to tackle the data scarcity issue of ERC by exploiting the unsupervised data. Specifically, we propose the Conversation Com- pletion (ConvCom) task based on unsupervised conversation data, which attempts to select the correct answer from candidate answers to fill a masked utterance in a conversation. Then, on the proposed ConvCom task, we Pre-train a basic COntext-Dependent Encoder (PRE-CODE). The hierarchical structure of the context-dependent en- coder makes our work different from …

Content truncated to 500 words