Automated Analysis of Mental Health Conversations Using Large Language Models and Sentiment Clustering
Xiangbo Zhang
Co-Presenters: Individual Presentation
College: The Dorothy and George Hennings College of Science, Mathematics and Technology
Major: Computer Science
Faculty Research Mentor: Malihe Aliasgari
Abstract:
Title: Automated Analysis of Mental Health Conversations Using Large LanguageModels and Sentiment ClusteringAuthors:Xiangbo Zhang, College of Science, Mathematics and Technology, Wenzhou-KeanUniversity, China;Parsa Safaee, Eng.D Candidate, Eindhoven University of Technology, Netherlands;Yousef Nejatbakhsh, Department of Mathematical Sciences, Kean University, USA;Malihe Aliasgari, School of Integrative Science and Technology, Kean University, USA.Abstract:Mental health conditions aGect millions worldwide, leading to communication diGiculties,behavioral changes, and emotional challenges that impact both individuals and caregivers.Traditional observation-based approaches place a significant burden on caregivers, whileexisting analytical tools often lack multi-language support and fail to capture emotionalnuances. This study presents an automated system that employs Large Language Models(LLMs) and OpenAI’s Whisper speech-to-text model to analyze conversations related tomental health. By extracting video and audio data from YouTube, the system processesmultilingual speech, applies sentiment analysis, and utilizes K-Means clustering to detectemotional patterns and behavioral trends. The findings indicate that sentiment polarity inmental health discussions is predominantly neutral to slightly positive, while clusteringtechniques reveal distinct emotional triggers and coping strategies. These insights supportthe development of empathy maps, providing data-driven guidance to enhancecommunication and emotional well-being. Future improvements include real-time analysisand cross-platform integration to broaden accessibility and eGectiveness in mental healthsupport.Keywords:Dementia, Large Language Models, Sentiment Analysis, Empathy Mapping, K-Means Clustering