Latent Suicide Risk Detection on Microblog via Suicide-Oriented Word Embeddings and Layered Attention

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 1718–1728, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 1718 Latent Suicide Risk Detection on Microblog via Suicide-Oriented Word Embeddings and Layered Attention Lei Cao1,2 and Huijun Zhang1,2 and Ling Feng1,2 and Zihan Wei3 Xin Wang1,2 and Ningyun Li1,2 and Xiaohao He1,2 1Department of Computer Science and Technology, Tsinghua University, Beijing, China 2Beijing National Research Center for Information Science and Technology, Beijing, China 3School of Software, Beihang University, Beijing, China {cao-l17,zhang-hj17,xin-wang18,liny18}@mails.tsinghua.edu.cn, fengling@mail.tsinghua.edu.cn, sy1721114@buaa.edu.cn, hexh17@mails.tsinghua.edu.cn Abstract Despite detection of suicidal ideation on social media has made great progress in recent years, people’s implicitly and anti-real contrarily ex- pressed posts still remain as an obstacle, con- straining the detectors to acquire higher satis- factory performance. Enlightened by the hid- den “tree holes” phenomenon on microblog, where people at suicide risk tend to disclose their inner real feelings and thoughts to the mi- croblog space whose authors have committed suicide, we explore the use of tree holes to en- hance microblog-based suicide risk detection from the following two perspectives. (1) We build suicide-oriented word embeddings based on tree hole contents to strength the sensibility of suicide-related lexicons and context based on tree hole contents. (2) A two-layered atten- tion mechanism is deployed to grasp intermit- tently changing points from individual’s open blog streams, revealing one’s inner emotional world more or less. Our experimental results show that with suicide-oriented word embed- dings and attention, microblog-based suicide risk detection can achieve over 91% accuracy. A large-scale well-labelled suicide data set is also reported in the paper. 1 Introduction Suicide is a growing problem in today’s society. Each year nearly 800,000 people worldwide com- mit suicide, which is one person every 40 seconds, and there are many more who attempt it (Organi- zation et al., 2014). Suicide prevention will con- duce to human’s well-being, of which timely sens- ing suicide ideation is an essential task. Existing Solutions. Traditional suicide risk as- sessment like Suicide Probability Scale (Bagge and Osman, 1998), Adult Suicide Ideation Ques- tionnaire (Fu et al., 2007), Suicidal Affect- Behavior-Cognition Scale (Harris et al., 2015), etc. requires respondents to either fill in a ques- Figure 1: An example of one user’s normal posts vs. his/her hidden tree hole posts on microblog. tionnaire or participate in a professional interview. However, they are applicable to a small group of people. Particularly for the people who are suffer- ing but tend to hide inmost thoughts and refuse to seek helps from others, the approaches can- not function (Essau, 2005; Rickwood et al., 2007; Zachrisson et al., 2006). Recently, the penetration of social media (like forums and microblogs) and its large-scale, low- cost, and open advantages enable researchers to overcome the limitation and timely detect individ- ual’s suicide ideation. Despite great efforts have been made (Alambo et al., 2019; Cheng et al., 2017; Du et al., 2018; Sawhney et al., 2018; Cop- …

Content truncated to 500 words