# A Computational Approach to Understanding Empathy Expressed in Text-Based Mental Health Support

**Authors:** Ashish Sharma, Adam S. Miner, David C. Atkins, Tim Althoff  
**arXiv:** [2009.08441](https://arxiv.org/abs/2009.08441)  
**Date:** 2020-09-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2009.08441v1/

## Key points

- Empathy is crucial in providing effective mental health support
- Measuring empathy in text-based platforms has been challenging
- A computational approach is proposed to understand how empathy is expressed online
- A novel framework for characterizing empathic communication in text-based conversations is developed
- A corpus of 10k annotated (post, response) pairs is created using the framework
- A multi-task RoBERTa-based bi-encoder model is introduced to identify empathy and extract rationales for predictions
- Experimental results show the effectiveness of the approach in identifying empathic conversations
- Users do not self-learn empathy over time in mental health interactions
- The finding emphasizes the need for empathy training and feedback within these platforms
- The research provides valuable insights into understanding and promoting empathy in text-based mental health support

## Summary

Empathy plays a crucial role in providing effective mental health support. Measuring empathy in text-based platforms has been challenging; however, this study proposes a computational approach to understanding how empathy is expressed in online mental health platforms. The authors develop a novel framework for characterizing empathic communication in text-based conversations and create a corpus of 10k annotated (post, response) pairs using this framework. They also introduce a multi-task RoBERTa-based bi-encoder model that can identify empathy in conversations and extract the underlying rationales for its predictions. Experimental results show the effectiveness of their approach in identifying empathic conversations. Furthermore, the authors analyze 235k mental health interactions and make an interesting observation that users do not self-learn empathy over time. This finding emphasizes the need for empathy training and feedback within these platforms. This research provides valuable insights into understanding and promoting empathy in text-based mental health support, offering opportunities to improve the quality of care provided through these platforms.