Automatic Suicide Identification and Proactive Crisis Management Based on Social Media

Abstract

Suicide is a great public health challenge. Two million people attempt suicide every year in China. Existing suicide prevention programs require the help-seeking initiative of suicidal individuals, but many of them have low motivation to seek help. The Internet has become an inseparable part of our life. However, very few researchers have investigated the impact of suicide-related social media use behaviors. Moreover, to our knowledge, no research has been conducted to test proactive and targeted preventions based on suicide-related social media use behaviors. To solve this problem and test the feasibility and acceptability of Proactive Suicide Prevention Online (PSPO), a new approach based on suicide-related online behaviors that combines proactive identification of suicidal individuals with specialized crisis management is needed. Firstly, we conducted a survey to identify suicide-related social media use behaviors that related to suicidal risk. Based on this study, we located a microblog group online. Next, those comments on a suicide note were analyzed by experts to provide a training set for machine learning models for suicide identification. Then the best-performing model was used to automatically identify posts that suggested suicidal thoughts and behaviors. Next, a microblog direct message containing crisis management information, including measurements covering suicide-related issues, depression, help-seeking behavior, model acceptability, and social media usage habit, was sent to suicidal users identified by the machine learning model. For those who replied to the message, trained counselors provided customized crisis management. Results showed that suicidal attempters showed significant higher level of suicidal ideation and more suicide-related social media use behaviours. Suicidal ideation affected suicidal attempt through the mediational chains of attended to suicidal information, commented on / reposted suicidal information or talked-about suicide online. Moreover, a total of 27,007 comments made in April 2017 were analyzed. Among these, 2,786 (10.32%) were classified as indicating suicidal thoughts and behaviors. The performance of the detection model was good. Among them, the best results of each model are as follows: accuracy (0.88), recall (0.85), F-measure (0.85), and accuracy (0.86). Between July 3, 2017 and July 3, 2018, we sent 24,727 direct messages to 12,486 social media users in total, and 5,542 (44.39%) responded. Over one third of the users who were contacted completed questionnaires included in the direct message. Out of the 1,403 valid responses, 1,259 participants (89.73%) reported suicidal ideation, but more than half (51.67%) reported that they had not sought help. The Patient Health Questionnaire-9 (PHQ-9) mean score was 17.40 (SD = 5.98). More than two thirds of the participants (968, 69.00%) thought the PSPO approach was acceptable. In addition, a total of 2,321 users replied to the direct message. In Comparison of the frequency of word usage in their microblog posts one-month before and one-month after the consultation, we found that the frequency of death-oriented words significantly declined while the frequency of future-oriented words significantly increased. The PSPO model which is based on well-established suicide-related social media use behaviors is suitable for identifying populations at-risk of suicide. Followed-up with proactive crisis management, it may be a useful supplement to existing prevention programs as it may increase the accessibility of anti-suicide information to people with suicidal thoughts and behaviors but low motivation to seek help.

Date of Award	17 Sept 2020
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Xiaonan Nancy YU (Supervisor) & Tingshao ZHU (External Supervisor)