Reproduce “Facilitating Multi-turn Emotional Support Conversation with Positive Emotion Elicitation: A Reinforcement Learning Approach” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jinfeng Zhou, Zhuang Chen, Bo Wang, Minlie Huang
Researchers from Tsinghua University developed SUPPORTER, a reinforcement learning framework with a mixture-of-expert architecture, designed to facilitate multi-turn emotional support conversations by explicitly focusing on positive emotion elicitation. The model dynamically balances empathy and elicitation, demonstrating superior performance in emotional support and dialogue diversity compared to baseline models, while maintaining conversational coherence.
Key insights
- Explicitly formalizing multi-turn emotional support conversation as a positive emotion elicitation process, rather than just empathy or strategy prediction, leads to more effective emotional guidance.
- A Mixture-of-Expert architecture, with specialized components for emotion and keyword processing, can effectively manage the complex and dynamic requirements of emotional support dialogues.
- Reinforcement learning, informed by a multi-faceted reward function that prioritizes both emotional transition and conversational quality, enables an agent to adaptively adjust its support intensity throughout an interaction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.