Social media content and suicidality: Implications for practice.

Area of Science:

• Psychology
• Computer Science
• Public Health

Background:

• Social media platforms like Instagram are increasingly used for personal expression, offering potential insights into mental health.
• Previous research on social media suicidality prediction primarily used text analysis and faced privacy challenges.
• A comprehensive approach integrating multimodal data (text, images, user features) is needed for accurate risk assessment.

Purpose of the Study:

• To develop and evaluate an AI model for predicting suicidal ideation risk using Instagram data from users in Hawaii.
• To explore the utility of visual and textual content, alongside user features, in identifying individuals at risk.
• To address privacy concerns by obtaining explicit user consent for data collection.

Main Methods:

• Collected online questionnaire data on suicidal ideation from 142 participants in Hawaii.
• Obtained consent to gather Instagram data (text, images, user features) for AI model training.
• Developed and validated an AI model to predict suicidal ideation based on collected multimodal data.

Main Results:

• The best AI model achieved 52% sensitivity, 92% specificity, 65% positive predictive value, and 82% accuracy in predicting suicidal ideation.
• Key predictors included negative descriptions, image characteristics (lower hue, more red), and specific content types (less nature/sky, more art/fashion/body parts/spoof).
• The study identified unique social-cultural influences from Hawaii impacting predictor relevance.

Conclusions:

• AI analysis of multimodal social media data holds significant potential for predicting suicidality risk.
• Findings can inform mental health providers in developing targeted online interventions and support strategies.
• Understanding context-specific predictors, like those in Hawaii, is crucial for effective suicide prevention gatekeeping on social media.