ESC-Skills: A Skill-Centric Framework for Emotional Support Conversations
Researchers have introduced ESC-Skills, a novel framework designed to identify and evolve emotional support capabilities for conversational agents. This system conceptualizes localized support interactions as Intervention Units (IUs), which encapsulate the dynamics of state-action-outcome among seeker states, support interventions, and emotional shifts following responses. An ESC-Skills Bank is created from both successful and unsuccessful dialogues, featuring guidance on interventions, conditions for applicability, anticipated outcomes, and associated risks. Additionally, a multi-profile self-evolutionary refinement framework enables an ESC agent to engage with various simulated profiles, enhancing its robustness. This method seeks to improve interpretability and facilitate systematic skill development compared to traditional end-to-end or broad strategy supervision techniques.
Key facts
- ESC-Skills is a skill-centric framework for emotional support conversations.
- It models interactions as Intervention Units (IUs) capturing state-action-outcome dynamics.
- IUs are extracted from both successful and failed ESC dialogues.
- The ESC-Skills Bank includes intervention guidance, conditions, outcomes, and risks.
- A multi-profile self-evolutionary refinement framework is introduced.
- The framework improves robustness through interaction with diverse simulated profiles.
- It addresses limitations of end-to-end response generation and coarse strategy supervision.
- The approach offers greater interpretability and systematic skill improvement.
Entities
—