Facilitating Multi-turn Emotional Support Conversation with Positive Emotion Elicitation: A Reinforcement Learning Approach

TL;DR

This paper introduces Supporter, a reinforcement learning model that enhances multi-turn emotional support conversations by explicitly focusing on eliciting positive emotions while maintaining coherence, improving emotional support effectiveness.

Contribution

It formalizes ESC as positive emotion elicitation and proposes a mixture-of-expert RL model with specialized rewards for emotion and coherence, advancing the field.

Findings

Supporter effectively elicits positive emotions during conversations.

The model maintains coherence while enhancing emotional support.

Experimental results show superior performance over baselines.

Abstract

Emotional support conversation (ESC) aims to provide emotional support (ES) to improve one's mental state. Existing works stay at fitting grounded responses and responding strategies (e.g., question), which ignore the effect on ES and lack explicit goals to guide emotional positive transition. To this end, we introduce a new paradigm to formalize multi-turn ESC as a process of positive emotion elicitation. Addressing this task requires finely adjusting the elicitation intensity in ES as the conversation progresses while maintaining conversational goals like coherence. In this paper, we propose Supporter, a mixture-of-expert-based reinforcement learning model, and well design ES and dialogue coherence rewards to guide policy's learning for responding. Experiments verify the superiority of Supporter in achieving positive emotion elicitation during responding while maintaining conversational goals including coherence.