A Privacy-Focused, Adaptable Chatbot for Detecting and Preventing Depression

According to the World Health Organization, approximately 280 million people have depression. However, mental health diagnostic and intervention tools remain largely inaccessible, unaffordable, and stigmatized. To address this global need, a novel, accessible, and nonintrusive diagnostic and assistive system was created through the development of two machine learning (ML) models and a web app. Psych2Go uses two ML models to nonintrusively detect depression (model 1) and emotion (model 2). Both achieve their respective goals by analyzing prosodic features in speech rather than the content. The first ML model achieves a depression detection accuracy of 75.54% and the second achieves an emotion detection accuracy of 77.60%. The assistive system is powered by the GPT-3.5 Turbo API. The API, using a custom prompt template, tailors responses and therapy techniques to the user-provided demographic information (name, gender, age) and the detected emotion from the second ML model. The prompt enables the GPT-3.5 Turbo API to apply cognitive behavioral therapy principles, identifying and addressing depression-related negative thoughts. Adhering to strict privacy standards, the chatbot eschews storage of personal conversations, focusing instead on session-specific data (the user’s time of a session, depression score, emotion, name, age, and gender). Psych2Go was deemed successful in providing privacy-focused, personalized emotional support and depression and emotion detection. The chatbot’s unprecedented privacy-focused approach and personalization allows for it to act as an aid for therapists to monitor progress and a support system for the user between therapy sessions.