Utilizing Large Language Models for Psychological Assessment: Enhancing Suicide Risk Detection Through Social Media Analysis

Mental health issues are increasingly prevalent due to the fast-paced lifestyle and rising work stress. Traditional psychological assessments face limitations due to resource scarcity and inefficiency. This paper presents a novel system using large language models (LLMs) to detect suicidal tendencies through social media analysis. Our system integrates prompt engineering and Retrieval-Augmented Generation (RAG) techniques to enhance detection accuracy. The architecture includes a knowledge retrieval-enhanced component and a judgment component, utilizing a fine-tuned LLM. Experiments with a Kaggle dataset demonstrate high performance, with consistent accuracy, precision, recall, and F1-Score across multiple runs. This system offers a promising solution to improve early psychological assessment and intervention, addressing the mental health resource gap. Future work aims to optimize the model and explore broader applications in psychological research.