Nanyang Technological University, Singapore, and the Institute of Mental Health have developed an AI framework that analyses casual conversations to identify young people at high risk of developing psychosis.
Unlike conventional AI models that attempt to predict mental health risks directly from speech transcripts, the new framework combines emotional signals extracted by large language models (LLMs) with traditional machine learning to assess psychosis risk, achieving up to 93% sensitivity in a study involving 400 participants.
Researchers developed LEMon-AID (Latent Emotion Proxy for Mental Health Detection with AI-integrated Decision-making), a framework that uses LLMs to assess a person's emotional state.
The LLMs extract subtle emotional profiles, including joy, trust, fear and sadness, from transcripts of young individuals' speech and combine these signals with predictions from traditional machine learning models.
LEMon-AID then generates risk assessments and explanations in natural language.
【MORE】