# Detecting and Explaining Crisis

**Authors:** Rohan Kshirsagar, Robert Morris, Sam Bowman  
**arXiv:** [1705.09585](https://arxiv.org/abs/1705.09585)  
**Date:** 2017-05-26  
**Summary page:** https://www.summarizepaper.com/en/paper/1705.09585v1/

## Key points

- Individuals on social media may reveal themselves to be in various states of crisis such as suicide, self-harm, abuse, or eating disorders.
- Detecting these crises automatically and accurately can have profound consequences for mental health professionals who can use these techniques to identify individuals who may be in need of support or intervention based on their online behavior.
- The authors propose an explanation as a coherent, concise subset of the text that rationalizes the crisis detection.
- To achieve this goal, they explore several methods to detect and explain crises using a combination of neural and non-neural techniques.
- The authors evaluate these techniques on a unique dataset obtained from Koko, an anonymous emotional support network available through various messaging applications.
- They annotate a small subset of the samples labeled with crisis with corresponding explanations.
- The authors find that their best technique significantly outperforms the baseline for both detection and explanation of crises in social media text.
- Overall, this study highlights the importance of combining machine learning techniques with human expertise to improve our understanding and response to mental health crises in digital spaces.

## Summary

In their paper titled "Detecting and Explaining Crisis," Rohan Kshirsagar, Robert Morris, and Sam Bowman explore the detection and explanation of crisis in social media text. They note that individuals on social media may reveal themselves to be in various states of crisis such as suicide, self-harm, abuse, or eating disorders. Detecting these crises automatically and accurately can have profound consequences for mental health professionals who can use these techniques to identify individuals who may be in need of support or intervention based on their online behavior. However, detecting a general state of crisis without explaining why has limited applications. Therefore, the authors propose an explanation as a coherent, concise subset of the text that rationalizes the crisis detection. To achieve this goal, they explore several methods to detect and explain crises using a combination of neural and non-neural techniques. The authors evaluate these techniques on a unique dataset obtained from Koko, an anonymous emotional support network available through various messaging applications. They annotate a small subset of the samples labeled with crisis with corresponding explanations. The authors find that their best technique significantly outperforms the baseline for both detection and explanation of crises in social media text. Overall, this study highlights the importance of combining machine learning techniques with human expertise to improve our understanding and response to mental health crises in digital spaces.