Towards Persona-Based Empathetic Conversational Models

Abstract

Empathetic conversational models have been shown to improve user satisfaction and task outcomes in numerous domains. In Psychology, persona has been shown to be highly correlated to personality, which in turn influences empathy. In addition, our empirical analysis also suggests that persona plays an important role in empathetic conversations. To this end, we propose a new task towards persona-based empathetic conversations and present the first empirical study on the impact of persona on empathetic responding. Specifically, we first present a novel large-scale multi-domain dataset for persona-based empathetic conversations. We then propose CoBERT, an efficient BERT-based response selection model that obtains the state-of-the-art performance on our dataset. Finally, we conduct extensive experiments to investigate the impact of persona on empathetic responding. Notably, our results show that persona improves empathetic responding more when CoBERT is trained on empathetic conversations than non-empathetic ones, establishing an empirical link between persona and empathy in human conversations.

Personalized Response Generation via Generative Split Memory Network

“…On the other hand, Mazare et al (2018) introduced personalization from Reddit (PCR) by incorporating the persona of each user with a (randomly chosen) subset of his/her posting comments. Zhong et al (2020) further extended their datasets with annotated empathy information (PEC). In this work, we combine those two different ways of gathering personalization signals of users, i.e., meta profile attributes and users' posting histories, and provide a more comprehensive, large scale personalized dataset derived from natural social conversations.…”

“…PCR denotes the persona-based dialog datasets from Reddit (Mazare et al, 2018). PEC denotes persona-based empathetic conversation (Zhong et al, 2020). PD denotes the PersonalDialog dataset (Zheng et al, 2020).…”

Proceedings of the Seventh Workshop on Computational Linguistics and Clinical Psychology: Improving Access

“…Our general conversation data is from two datasets: Persona-based Empathetic Conversation (PEC) (Zhong et al, 2020) and EmpatheticDialogues (ED) (Rashkin et al, 2019). Their statistics are listed in Table 1.…”

“…Our models have no access to counselling conversations during their training and validation, as we experiment with learning from contrast of empathy vs. non-empathy in out-of-domain (OOD) training data. To that end, we leverage publicly available datasets of general conversations with heuristic empathy labels (Rashkin et al, 2019;Zhong et al, 2020) for OOD training, investigating the connections between general-conversational empathy and therapeutic empathy, as illustrated in Figure 1.…”

Towards Low-Resource Real-Time Assessment of Empathy in Counselling

“…Our models have no access to counselling conversations during their training and validation, as we experiment with learning from contrast of empathy vs. non-empathy in out-of-domain (OOD) training data. To that end, we leverage publicly available datasets of general conversations with heuristic empathy labels (Rashkin et al, 2019;Zhong et al, 2020) for OOD training, investigating the connections between general-conversational empathy and therapeutic empathy, as illustrated in Figure 1.…”