COSMIC: COmmonSense knowledge for eMotion Identification in Conversations

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 2470–2481 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 2470 COSMIC: COmmonSense knowledge for eMotion Identification in Conversations Deepanway Ghosal†, Navonil Majumder†, Alexander Gelbukh⋄, Rada Mihalcea△, Soujanya Poria† † Singapore University of Technology and Design, Singapore ⋄ CIC, Instituto Polit´ecnico Nacional, Mexico △ University of Michigan, USA {deepanway ghosal@mymail.,navonil majumder@, sporia@}sutd.edu.sg, gelbukh@cic.ipn.mx, mihalcea@umich.edu Abstract In this paper, we address the task of utter- ance level emotion recognition in conversa- tions using commonsense knowledge. We propose COSMIC, a new framework that incorporates different elements of common- sense such as mental states, events, and causal relations, and build upon them to learn interactions between interlocutors participat- ing in a conversation. Current state-of-the- art methods often encounter difficulties in context propagation, emotion shift detection, and differentiating between related emotion classes. By learning distinct commonsense representations, COSMIC addresses these chal- lenges and achieves new state-of-the-art re- sults for emotion recognition on four differ- ent benchmark conversational datasets. Our code is available at https://github.com/ declare-lab/conv-emotion. 1 Introduction Emotion recognition is a long-standing research problem in Artificial Intelligence (AI). With the growing popularity of conversational AI research, the topic of emotion recognition in conversations has received significant attention from the research community (Li et al., 2020; Ghosal et al., 2019; Zhang et al., 2019). Identifying emotions in con- versations is a core step toward fine-grained con- versation understanding, which in turn is essen- tial for downstream tasks such as emotion-aware chat agents (Lin et al., 2019; Rashkin et al., 2019), visual question answering (Tapaswi et al., 2016; Azab, 2019), health conversations (Althoff et al., 2016; P´erez-Rosas et al., 2017) and others. Natural conversations are complex as they are governed by several distinct variables that affect the flow of a conversation and the emotional dy- namics of the participants. These variables include Reaction of A: Gets tired Reaction of B: Irritated Effect on B: Gets yelled at Angry Angry Person A Person B Look, it's a beautiful day outside, why are we arguing? Well, what do you want me to do about it? What do you want? Reaction of A: Angry, annoyed Intent of B: Help out Effect on B: Thinks what to do Commonsense Inference Commonsense Inference I want you to pretend like he's coming back. Angry Influenced by the other person Figure 1: Commonsense knowledge can lead to ex- plainable dialogue understanding. It will help mod- els to understand, reason, and explain events and sit- uations. In this particular example, commonsense in- ference is applied to a sequence of utterances in a two- party conversation. Person A’s first utterance indicates that he/she is tired of arguing with person B. The tone of the utterance also implies that person B is getting yelled at by person A, which invokes a reaction of ir- ritation in person B. Person B then asks what he/she can do to help and says this while being angry. This again makes person A annoyed and influences him/her to respond …

Content truncated to 500 words