Self-Adapted Utterance Selection for Suicidal Ideation Detection in Lifeline Conversations

Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 1436–1446 May 2-6, 2023 ©2023 Association for Computational Linguistics Self-Adapted Utterance Selection for Suicidal Ideation Detection in Lifeline Conversations Zhong-Ling Wang Department of Computer Science National Chengchi University Taipei, Taiwan 109753106@g.nccu.edu.tw Po-Hsien Huang and Wen-Yau Hsu Department of Psychology National Chengchi University Taipei, Taiwan psyphh@nccu.edu.tw hsu@nccu.edu.tw Hen-Hsen Huang Institute of Information Science Academia Sinica Taipei, Taiwan hhhuang@iis.sinica.edu.tw Abstract This paper investigates a crucial aspect of men- tal health by exploring the detection of suici- dal ideation in spoken phone conversations be- tween callers and counselors at a suicide pre- vention hotline. These conversations can be lengthy, noisy, and cover a broad range of top- ics, making it challenging for NLP models to accurately identify the caller’s suicidal ideation. To address these difficulties, we introduce a novel, self-adaptive approach that identifies the most critical utterances that the NLP model can more easily distinguish. The experiments use real-world Lifeline transcriptions, expertly la- beled, and show that our approach outperforms the baseline models in overall performance with an F-score of 66.01%. In detecting the most dangerous cases, our approach achieves a significantly higher F-score of 65.94% com- pared to the baseline models, an improvement of 8.9%. The selected utterances can also pro- vide valuable insights for suicide prevention research. Furthermore, our approach demon- strates its versatility by showing its effective- ness in sentiment analysis, making it a valuable tool for NLP applications beyond the health- care domain. 1 Introduction The suicide prevention hotline provides a free ser- vice for the people with mental issues to have the support on a 24/7 basis. On the other end of the call, a group of counselors who are trained vol- unteers are waiting for calls online in shifts. As mental health problems are becoming more and more prevalent in society, the difficulty of finding a sufficient number of counselors forms a com- mon challenge that the suicide prevention hotline of many countries in the world is facing. The over- whelmed counselors and the busy phone line make the individuals in urgent need of help unable to get the opportunity to have the counseling. In order to alleviate of demand of suicide counseling, the intelligent conversation system forms a potential direction to study for aiding the hotline service. Fully automated chatbots as counselors providing counseling services are still impractical because the current natural language processing (NLP) technol- ogy has not yet been able to construct a reliable chatbot like a well-trained counselor to talk to these people who are mentally unstable and at high risk of suicide. In this work, we approach the suicide counsel- ing aiding from a more practical aspect. Instead of creating a chatbot to replace the human counselors, our goal is to propose a model for suicidal ideation detection (Ji et al., 2021). During the counseling, our model can be used for monitoring the conver- sation and identifying the suicide attempt of the caller. Once it is …

Content truncated to 500 words