Physiology-Driven Empathic Large Language Models (EmLLMs) for Mental Health Support

Abstract

Wearable devices show promise in monitoring and managing mental health, but gaps exist in accurately predicting users' mental states and cognitively engaging with users to provide mental health support with wearable data. In this proposal, I present the concept of physiology-driven Empathic Large Language Models (EmLLMs) for mental health support. EmLLMs monitor users and their surrounding environment using wearable devices to predict their mental and emotional states and interact with them based on these states. I present the application of this approach for monitoring and managing excess stress in the workplace. To improve the accuracy of stress prediction, I developed a novel Science-Guided Machine Learning (SGML) model that automatically extracts features from raw wearable data. To engage with users cognitively, I developed an (EmLLM) chatbot that provides psychotherapy based on predicted user stress. I present the SGML model's preliminary findings and results from a pilot user study that evaluates the EmLLM chatbot.