Knowledge-enhanced Mixed-initiative Dialogue System for Emotional Support Conversations

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 4079–4095 July 9-14, 2023 ©2023 Association for Computational Linguistics Knowledge-enhanced Mixed-initiative Dialogue System for Emotional Support Conversations Yang Deng1, Wenxuan Zhang2,†, Yifei Yuan1, Wai Lam1 1 The Chinese University of Hong Kong, 2 DAMO Academy, Alibaba Group {dengyang17dydy,isakzhang}@gmail.com {yfyuan,wlam}@se.cuhk.edu.hk Abstract Unlike empathetic dialogues, the system in emotional support conversations (ESC) is ex- pected to not only convey empathy for comfort- ing the help-seeker, but also proactively assist in exploring and addressing their problems dur- ing the conversation. In this work, we study the problem of mixed-initiative ESC where the user and system can both take the initiative in leading the conversation. Specifically, we conduct a novel analysis on mixed-initiative ESC systems with a tailor-designed schema that divides utterances into different types with speaker roles and initiative types. Four emo- tional support metrics are proposed to evaluate the mixed-initiative interactions. The analysis reveals the necessity and challenges of building mixed-initiative ESC systems. In the light of this, we propose a knowledge-enhanced mixed- initiative framework (KEMI) for ESC, which retrieves actual case knowledge from a large- scale mental health knowledge graph for gener- ating mixed-initiative responses. Experimental results on two ESC datasets show the superi- ority of KEMI in both content-preserving eval- uation and mixed initiative related analyses. 1 Introduction As the world is making efforts to recover from Covid-19 and plans for future construction, emo- tional support is of great importance in resolving the widespread emotional distress and increased risk for psychiatric illness associated with the pan- demic (Pfefferbaum and North, 2020; Suh et al., 2021). A wide range of emotional support con- versation (ESC) systems are emerging to provide prompt and convenient emotional support for help- seekers, including mental health support (Sharma ∗ The work described in this paper is substantially sup- ported by a grant from the Research Grant Council of the Hong Kong Special Administrative Region, China (Project Code: 14200620). † Corresponding author. 😞 I lost my job last year and got really angry. 😡 I am sorry to hear that. It must be so upset. (Non-Initiative) Yes, it was a complete surprise. 😡 🤖 I can understand that. I hope it turned out to be a blessing. (Non-Initiative) I wish I had another job today. 😡 🤖 I feel your pain. It happened to me few years ago. I hope you find a job, good luck! (Non-Initiative) I'm in depression because recently I lost my job. I am so sorry to hear that. Did you work there for a long time? (Initiative) 😞 Yes up to five years but my company getting too much lost so they took out lots of employs. 🤖 I can understand. It can cause a lot of depression for you. (Non-Initiative) 😞 I don’t know what to do now. I would recommend looking for a local group that help assist creating new resumes. Nonprofits have people who volunteer. (Initiative) EmpatheticDialogues ESConv 🤔 🤔 🤖 Figure …

Content truncated to 500 words