A Time-Aware Transformer Based Model for Suicide Ideation Detection on Social Media

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 7685–7697, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 7685 A Time-Aware Transformer Based Model for Suicide Ideation Detection on Social Media Ramit Sawhney Netaji Subhas Institute of Technology ramits.co@nsit.net.in Harshit Joshi University of Delhi harshit113@ducic.ac.in Saumya Gandhi Visvesvaraya National Institute of Technology gandhisaumya8@gmail.com Rajiv Ratn Shah IIIT Delhi rajivratn@iiitd.ac.in Abstract Social media’s ubiquity fosters a space for users to exhibit suicidal thoughts outside of tra- ditional clinical settings. Understanding the build-up of such ideation is critical for the identification of at-risk users and suicide pre- vention. Suicide ideation is often linked to a history of mental depression. The emotional spectrum of a user’s historical activity on so- cial media can be indicative of their mental state over time. In this work, we focus on identifying suicidal intent in English tweets by augmenting linguistic models with histori- cal context. We propose STATENet, a time- aware transformer based model for prelimi- nary screening of suicidal risk on social media. STATENet outperforms competitive methods, demonstrating the utility of emotional and tem- poral contextual cues for suicide risk assess- ment. We discuss the empirical, qualitative, practical, and ethical aspects of STATENet for suicide ideation detection.1 1 Introduction Globally, close to 800,000 people die by suicide each year, and 20 times more people attempt sui- cide. Suicide is the second leading cause of death in the 15 to 29 year age group (WHO, 2014) with a rising suicide rate of 35% in the US since 1999 (Hedegaard et al., 2020). Extending clinical and psychological care to people showing suicidal ideation relies heavily on identifying those at risk. Tragically, 80% of patients do not undergo psychi- atric treatment, and about 60% of those who died of suicide denied having suicidal thoughts to mental health practitioners (McHugh et al., 2019). Recent studies (Coppersmith et al., 2018) also show that people exhibiting suicidal ideation make frequent use of social media, e.g., Twitter, to share their 1https://github.com/midas-research/ STATENet_Time_Aware_Suicide_Assessment josh is so cute Hiiii belated merry christmas and advanced happy new year buddy My friends tell me its difficult to trust me since I keep joking all the time I do not want to be alive, I just want to die today. Please kill me already. If someone does that I will be happy 27/08/2011 28/08/2011 08/09/2011 12/10/2014 12/10/2014 ~ 3 years Tweet to assess A user’s tweeting history Tweet #1 Tweet #2 Tweet #3 Tweet #4 No suicidal or depressive tendencies Concerning, clear suicidal intent Figure 1: We study a user whose latest tweet is not in- dicative of suicidal intent. Without seeing the user’s re- cent historic tweet, which shows self-harm tendencies, it is difficult to accurately assess suicidal risk. How- ever, analyzing a user’s tweeting history sequentially without factoring in time irregularities between tweets may lead to an inaccurate representation of a user’s mental state. Time-aware modeling of the temporal de- pendency between historic tweets reduces the impact of tweets from 3 years …

Content truncated to 500 words