Predicting Suicide Using Natural Language Processing of Autobiographical Memory

Autobiographical memory, a critical cognitive process for recalling personal events, is closely linked to mental health. Depressive disorders are characterized by overgeneralized and negative memory patterns, which impair future-oriented thinking and exacerbate hopelessness. Current evaluations of autobiographical memory are subjective and limited by human bias. In this study, we applied Natural Language Processing using Large Language Models (LLMs) to analyze autobiographical memory narratives, uncovering that their valence can predict depression, suicidal ideation, and prior suicide attempts. Furthermore, valence correlated with core components of the Three-Step Theory of suicide, such as hopelessness and lack of connectedness. By integrating advanced computational techniques, our approach demonstrated high predictive accuracy and offers a scalable, objective method for assessing suicide risk. These findings highlight the potential of LLM-based analysis in enhancing psychological assessment and informing interventions, paving the way for innovative clinical applications in mental health care.