Towards Suicide Ideation Detection Through Online Conversational Context Ramit Sawhney ∗2 rsawhney31@gatech.edu Georgia Institute of Technology USA Shivam Agarwal ∗2 shivama2@illinois.edu University of Illinois at Urbana-Champaign USA Atula Tejaswi Neerkaje ∗ atulatejaswi@gmail.com Conversational AI and Social Analytics (CAISA) Lab, University of Marburg Germany Nikolaos Aletras n.aletras@shefeld.ac.uk University of Shefeld United Kingdom Preslav Nakov pnakov@hbku.edu.qa Qatar Computing Research Institute, HBKU Doha, Qatar Lucie Flek lucie.fek@uni-marburg.de Conversational AI and Social Analytics (CAISA) Lab, University of Marburg Germany ABSTRACT Social media enable users to share their feelings and emotional struggles. They also ofer an opportunity to provide community support to suicidal users. Recent studies on suicide risk assessment have explored the user’s historic timeline and information from their social network to analyze their emotional state. However, such methods often require a large amount of user-centric data. A less intrusive alternative is to only use conversation trees arising from online community responses. Modeling such online conversations between the community and a person in distress is an important context for understanding that person’s mental state. However, this is not trivial since comments can have diverse infuence on a user in distress. Typically, a handful of posts receive a signifcantly high number of replies, which results in scale-free dynamics in the conversation tree. Moreover, psychological studies suggested that it is important to capture the fne-grained temporal irregularities in the release of vast volumes of comments, since suicidal users react quickly to online community support. Building on these limitations and psychological studies, we propose HCN, a Hyperbolic Conver- sation Network, a less user-intrusive method for suicide ideation detection. Through extensive quantitative, qualitative, and ablative experiments on real-world Twitter data, we fnd that HCN outper- forms state-of-the art methods, while using 98% less user-specifc data, and while maintaining a 74% lower carbon footprint and a 94% smaller model size. We also fnd that the comments within the frst half an hour are most important to identify at-risk users. ∗ Authors contributed equally. 2 Also with Conversational AI and Social Analytics (CAISA) Lab, University of Marburg. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for proft or commercial advantage and that copies bear this notice and the full citation on the frst page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specifc permission and/or a fee. Request permissions from permissions@acm.org. SIGIR ’22, July 11ś15, 2022, Madrid, Spain © 2022 Association for Computing Machinery. ACM ISBN 978-1-4503-8732-3/22/07. . . $15.00 https://doi.org/10.1145/3477495.3532068 CCS CONCEPTS · Computing methodologies → Natural language processing; · Human-centered computing → Social media. KEYWORDS suicide ideation, conversation trees, social media ACM Reference Format: Ramit Sawhney, Shivam Agarwal, Atula Tejaswi Neerkaje, Nikolaos Aletras, Preslav Nakov, and Lucie Flek. 2022. Towards Suicide Ideation Detection Through Online Conversational Context. In Proceedings of the 45th Interna- tional ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’22), July 11ś15, 2022, Madrid, Spain. ACM, New York, NY, USA, 12 pages. https://doi.org/10.1145/3477495.3532068 1 INTRODUCTION Social media enable people with suicidal thoughts to share their mental struggles with a large number of users online [84]. As a result, the connection between social media and suicide has become a public health concern [65], e.g., a study has shown that eight out of ten people who exhibit suicide ideation tend to disclose their suicidal thoughts online [39]. As a result, research eforts have in- creasingly shifted to analyzing mental health using social media for identifying individuals at risk [53], with the aim of potentially extending appropriate care [22, 113]. Recent advances in computa- tional social science have made progress in assessing the suicide risk on the web, e.g., using context-enriched models that leverage the user’s posting history [69, 90] or their social network connec- tions [91]. However, such methods typically retrieve large volumes of user-specifc data and require intensive computational resources. Psychological research has shown that in order to properly assess a social media post, it is crucial to study the nature of the conversa- tion fow on contemplating suicide and also to understand the level of community support ofered to suicidal individuals [26, 94, 101]. Studies have shown that online conversations can highly infuence the mental states of online users [51, 97]. Online peers often ofer support to users in distress and discourage them from taking the extreme step, but are sometimes dismissive and pro-suicidal [78]. Topic 23: Social Aspects SIGIR ’22, July 11–15, 2022, Madrid, Spain 1716