MIT News breed · Science

AI Tool Estimates Suicide Risk from Text

Scientists developed a language tool analyzing text messages to estimate suicide risk, using a lexicon of words linked to 49 risk factors.

The MIT tool analyzes text conversations from Crisis Text Line, aiming to identify key suicide risk factors during mental health crises.

Suicide prediction is challenging due to numerous interacting risk factors. Researchers focused on real-time indicators during crises.

A suicide-risk lexicon was built using AI and expert clinicians, containing terms for 49 risk factors.

A machine learning model trained on this lexicon found mentions of lethal means and substance use were strong risk predictors.

The model assigns weights to risk factors, with lethal means being heavily weighted. It accurately predicted risk severity on new conversations.

Limitations include context insensitivity. The tool is a simpler, interpretable model, addressing computational and privacy concerns. Human oversight is critical, and validation is needed for clinical use.

AI-samenvatting op basis van de bron.