Combining Psychological Theory with Language Models for Suicide Risk Detection

Findings of the Association for Computational Linguistics: EACL 2023, pages 2430–2438 May 2-6, 2023 ©2023 Association for Computational Linguistics Combining Psychological Theory with Language Models for Suicide Risk Detection Daniel Izmaylov1 izmaylov@post.bgu.ac.il Amir Bialer1 amirbial@post.bgu.ac.il Avi Segal1 avisegal@gmail.com Meytal Grimland2 meytal.grimland@gmail.com Yossi Levi-Belz2 yossil@ruppin.ac.il Kobi Gal1,3 kobig@bgu.ac.i 1Ben-Gurion University of the Negev 2Ruppin Academic Center 3University of Edinburgh Abstract Recent years saw a dramatic increase in the pop- ularity of online counseling services providing emergency mental health support. This paper provides a new language model for automatic detection of suicide risk in online chat sessions between help-seekers and counselors. The model adapts a hierarchical BERT language model for this task. It extends the state of the art in capturing aspects of the conversation struc- ture in the counseling session and in integrating psychological theory into the model. We test the performance of our approach in a leading national online counseling service that operates in the Hebrew language. Our model outper- formed other non-hierarchical approaches from the literature, achieving a 0.76 F2 score and 0.92 ROC-AUC. Moreover, we demonstrate our model’s superiority over strong baselines even early on in the conversation, which is key for real-time detection in the field. This is a first step towards incorporating suicide predictive models in online support services and advanc- ing NLP tools for resource-bounded languages. 1 Introduction Suicide accounts for more than 700,000 lives lost across the world every year. It is the second leading cause of death for adolescents and adults from 15 to 29 years of age in many countries. A key effort in suicide prevention is to identify individuals at risk of suicide as early as possible (World-Health- Organization, 2021). In the past decade, online counseling services for mental health support have become common- place in many countries, providing chat support and guidance to at-risk individuals (see fictitious example in Figure 1). Online counseling services aim to provide mental support and address a va- riety of mental health crises through specialist counselors. These counselors are trained to de- tect suicide risk during conversations and inter- vene quickly as needed. These services have ex- Figure 1: A fictitious example of a conversation. perienced tremendous growth in traffic since the commencement of the COVID pandemic (Zalsman et al., 2021). Any kind of technological support to help counselors in this critical task can potentially save lives. This paper provides a computational model for detection of suicide risk from anonymous text- based discussions between help-seekers and coun- selors. Our data is taken from an online counsel- ing service in a low-resource language (Hebrew). There are several challenges towards solving sui- cide risk detection in our setting: State of the art pre-trained language models for suicide prevention usually focus on posts from social media which are very different in structure from online conver- sations between counselors and help-seekers. Ex- isting works that do consider conversations in this domain ignore the conversation structure or are limited in the size of the conversation they con- sider. Also, the …

Content truncated to 500 words