# Towards Facilitating Empathic Conversations in Online Mental Health Support: A Reinforcement Learning Approach

**Authors:** Ashish Sharma, Inna W. Lin, Adam S. Miner, David C. Atkins, Tim Althoff  
**arXiv:** [2101.07714](https://arxiv.org/abs/2101.07714)  
**Date:** 2021-01-19  
**Summary page:** https://www.summarizepaper.com/en/paper/2101.07714v3/

## Key points

- Online peer-to-peer support platforms can improve access to mental health treatment and reduce disease burden.
- Highly empathic conversations are rare on these platforms.
- The authors propose a new approach called PARTNER to increase empathy in online mental health support conversations.
- PARTNER is a deep reinforcement learning agent that makes sentence-level edits to increase empathy while maintaining conversation quality.
- The agent uses a policy network based on a transformer language model adapted from GPT-2.
- During training, transformations that increase empathy while maintaining text fluency, context specificity, and diversity are rewarded.
- PARTNER generates more empathic, specific, and diverse responses compared to NLP methods from related tasks like style transfer and empathic dialogue generation.
- Improving empathy in online mental health support conversations can better meet the needs of individuals seeking mental health support and contribute to improved treatment outcomes globally.

## Summary

Online peer-to-peer support platforms have the potential to greatly improve access to mental health treatment and reduce the global burden of disease. Recent studies have shown that highly empathic conversations are rare in these platforms. To address this issue, the authors propose a new approach called PARTNER which aims to increase empathy in online mental health support conversations. PARTNER is a deep reinforcement learning agent that learns to make sentence-level edits to posts in order to increase the expressed level of empathy while maintaining conversation quality. The agent leverages a policy network based on a transformer language model adapted from GPT-2. This network performs the dual task of generating candidate empathic sentences and adding them at appropriate positions within the conversation. During training, transformations that increase empathy in posts while maintaining text fluency, context specificity and diversity are rewarded. Through a combination of automatic and human evaluation it is demonstrated that PARTNER successfully generates more empathic, specific and diverse responses compared to NLP methods from related tasks like style transfer and empathic dialogue generation. The findings of this study have direct implications for facilitating empathic conversations on web-based platforms. By improving empathy in online mental health support conversations these platforms can better meet the needs of individuals seeking mental health support and potentially contribute to improved treatment outcomes and reduced disease burden globally.