New AI-Driven Tools for Enhancing Campus Well-being: A Prevention and Intervention Approach
A new dissertation introduces an integrated AI framework to improve campus well-being through prevention and intervention strategies. Addressing the lack of effective mental health monitoring in universities, the research presents TigerGPT, a personalized survey chatbot using Large Language Models (LLMs) that achieved 81% user satisfaction. To enhance conversation depth, the AURA reinforcement-learning framework was developed, adapting follow-up questions based on quality signals and significantly improving interaction metrics. For intervention, the study analyzes Expressive Narrative Stories using BERT models to detect nuanced linguistic features without relying on explicit keywords. Additionally, PsychoGPT, an LLM aligned with DSM-5 and PHQ-8 guidelines, performs distress classification and symptom scoring. To mitigate hallucinations, the Stacked Multi-Model Reasoning (SMMR) technique layers expert models, outperforming single-model solutions in accuracy and F1 scores on the DAIC-WOZ dataset. This cohesive system allows adaptive survey insights to directly inform specialized mental health detection models, offering a comprehensive solution for university student support services.
Wire timeline
New AI-Driven Tools for Enhancing Campus Well-being: A Prevention and Intervention Approach
A new dissertation introduces an integrated AI framework to improve campus well-being through prevention and intervention strategies. Addressing the lack of effective mental health monitoring in universities, the research presents TigerGPT, a personalized survey chatbot using Large Language Models (LLMs) that achieved 81% user satisfaction. To enhance conversation depth, the AURA reinforcement-learning framework was developed, adapting follow-up questions based on quality signals and significantly improving interaction metrics. For intervention, the study analyzes Expressive Narrative Stories using BERT models to detect nuanced linguistic features without relying on explicit keywords. Additionally, PsychoGPT, an LLM aligned with DSM-5 and PHQ-8 guidelines, performs distress classification and symptom scoring. To mitigate hallucinations, the Stacked Multi-Model Reasoning (SMMR) technique layers expert models, outperforming single-model solutions in accuracy and F1 scores on the DAIC-WOZ dataset. This cohesive system allows adaptive survey insights to directly inform specialized mental health detection models, offering a comprehensive solution for university student support services.
cs.AI updates on arXiv.org