Multi-task learning to detect suicide ideation and mental disorders among social media users

Area of Science:

• Computational psychiatry
• Social media analytics
• Artificial Intelligence in healthcare

Background:

• Mental disorders and suicide pose significant global health challenges.
• Early detection of mental illness and suicide ideation is crucial for intervention.
• Social media data offers a potential avenue for large-scale mental health monitoring.

Purpose of the Study:

• To investigate the effectiveness of using shared representations for detecting mental illness and suicide ideation from social media.
• To explore the impact of comorbidity on suicide ideation prediction.
• To validate the generalizability and accuracy of AI models across different social media platforms.

Main Methods:

• Utilized multi-task learning (MTL) with soft and hard parameter sharing.
• Extracted shared features between mental illness and suicide ideation detection tasks.
• Employed cross-platform knowledge sharing and predefined auxiliary inputs.
• Tested model generalizability using two distinct social media datasets.

Main Results:

• Achieved state-of-the-art results in detecting users with suicide ideation.
• Demonstrated increased predictive accuracy for suicide risk using data from individuals with comorbid mental disorders.
• Identified specific mental disorders, such as Post-Traumatic Stress Disorder (PTSD), with a noticeable impact on suicidal risk.
• Validated the effectiveness of cross-platform knowledge sharing for improved prediction.

Conclusions:

• AI, particularly MTL, can effectively detect mental illness and suicide ideation from social media data.
• Comorbidity significantly enhances the accuracy of suicide risk prediction.
• Cross-platform data integration and auxiliary inputs further boost model performance.
• Findings support the development of AI-driven tools for proactive mental healthcare.