Visualizing Mental Health Insights: A Pipeline from Social Media to Chernoff Faces.

This study proposes an approach for analyzing mental health through publicly available social media data, employing Large Language Models (LLMs) and visualization techniques to transform textual data into Chernoff Faces. The analysis began with a dataset comprising 15,744 posts sourced from major social media platforms, which was refined down to 2,621 posts through meticulous data cleaning, feature extraction, and visualization processes. Our methodology includes stages of Data Preparation, Feature Extraction, Chernoff Face Visualization, and Clinical Validation. Dimensionality reduction techniques such as PCA, t-SNE, and UMAP were employed to transform complex mental health data into comprehensible visual representations. Validation involved a survey among 60 volunteer psychiatrists, underscoring the visualizations' potential for enhancing clinical assessments. This work sets the stage for future evaluations, specifically focusing on a combined features method to further refine the visual representation of mental health conditions and to augment the diagnostic tools available to mental health professionals.