Empathetic Reflective Response Generation: Towards Conversation Models for Online Mental Health Support

This paper presents an empathetic generative model that responds to online mental-health group dialogues with high-quality counseling-style reflection. Reflection is a crucial counseling technique that involves actively listening to the inter-locutor's thoughts and emotions, interpreting their meaning, and then articulating a concise, empathetic response to demonstrate understanding and promote deeper self-exploration. We discuss the importance of empathy in effective counseling approaches and study the impact of modeling multiple text-based empathetic factors on generating better reflective responses. Our model inte-grates various text-based empathetic components such as emotion and intent to enhance context and response representations, as well as three mechanisms for the communication of empathy in response. Empirical experiments demonstrate the significance of our approach through automated and human evaluations in terms of reflectiveness, empathy, coherence, and fluency.