Social Media Content and Suicidality: Implications for Practice

Abstract

Artificial intelligence is a useful tool for examining suicidality on social media, where people share their thoughts. However, existing research has primarily focused on text analysis to predict risk in single posts and raised privacy concerns. This study aimed to use text, images, and user features on Instagram to predict the risk of each user in Hawaii with user permission. One hundred forty-two participants completed online questionnaires about their current suicidal ideation. With their consent, their Instagram data were collected to train Artificial intelligence model to predict suicidal ideation of each user. Thirty-three (23.2%) participants reported having current suicidal ideation. The best model predicts suicidal ideation with 52% sensitivity, 92% specificity, 65% positive predictive value, and 82% accuracy. The significant predictors were negative description, lower hue and more red in an image, fewer nature and sky images, more art, fashion, a close-up of a body part, and spoof content in an image. These findings highlight the potential of suicide prediction on social media, which help mental health providers plan patient online interventions and appointments. Additionally, the influence of Hawaii’s unique social-cultural context on significant predictors, helping gatekeepers to recognize signs of suicide on the social media of people in Hawaii.