CoMPM: Context Modeling with Speaker’s Pre-trained Memory Tracking for Emotion Recognition in Conversation

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 5669 - 5679 July 10-15, 2022 ©2022 Association for Computational Linguistics CoMPM: Context Modeling with Speaker’s Pre-trained Memory Tracking for Emotion Recognition in Conversation Joosung Lee, Wooin Lee Kakao Enterprise Corp., South Korea {rung.joo, dan.kes}@kakaoenterprise.com Abstract As the use of interactive machines grow, the task of Emotion Recognition in Conversa- tion (ERC) became more important. If the machine-generated sentences reflect emotion, more human-like sympathetic conversations are possible. Since emotion recognition in con- versation is inaccurate if the previous utterances are not taken into account, many studies re- flect the dialogue context to improve the per- formances. Many recent approaches show per- formance improvement by combining knowl- edge into modules learned from external struc- tured data. However, structured data is difficult to access in non-English languages, making it difficult to extend to other languages. There- fore, we extract the pre-trained memory using the pre-trained language model as an extractor of external knowledge. We introduce CoMPM, which combines the speaker’s pre-trained mem- ory with the context model, and find that the pre-trained memory significantly improves the performance of the context model. CoMPM achieves the first or second performance on all data and is state-of-the-art among systems that do not leverage structured data. In addition, our method shows that it can be extended to other languages because structured knowledge is not required, unlike previous methods. Our code is available on github 1. 1 Introduction As the number of applications such as interactive chatbots or social media that are used by many users has recently increased dramatically, Emotion Recognition in Conversation (ERC) plays a more important role in natural language processing, and as a proof, a lot of research (Poria et al., 2019; Zhang et al., 2019; Ghosal et al., 2020; Jiao et al., 2020) has been conducted on the task. The ERC module increases the quality of em- pathetic conversations with the users and can be 1https://github.com/rungjoo/CoMPM Figure 1: An example of MELD dataset utilized when sending tailored push messages to the users (Shin et al., 2019; Zandie and Mahoor, 2020; Lin et al., 2020). In addition, emotion recognition can be effectively used for opinion mining, rec- ommender systems, and healthcare systems where it can improve the service qualities by providing personalized results. As these interactive machines increase, the ERC module plays an increasingly important role. Figure 1 is an example of a conversation in which two speakers are angry at each other. The emotion of speaker B’s utterance ("How’d you get to that?") is angry. If the system does not take into account previous utterances, it is difficult to properly recog- nize emotions. Like the previous studies (Ghosal et al., 2020), we show that the utterance-level emo- tion recognition, which does not consider the pre- vious utterance, have limitations and experiments result in poor performances. Therefore, recent studies are attempting to recognize emotions while taking into account the previous utterances. Representatively, Dia- …

Content truncated to 500 words