Automated Detection and Classification of Delusion-related Content in Naturalistic Audio Diaries Using Multi-Agent Language Models
Researchers have developed a novel multi-agent language model pipeline to automatically detect and classify delusion-related content in audio diaries. The system, evaluated on transcripts from individuals with persecutory ideation, demonstrated robust performance using a majority voting framework, achieving a Micro F1 score of 0.872 for delusion detection and 0.779 for classification. This approach offers a scalable method for analyzing speech to identify and characterize content suggestive of delusional beliefs. AI
IMPACT Provides a scalable method for automated analysis of speech to identify and characterize content suggestive of delusional beliefs.