First insights on a passive major depressive disorder prediction system with incorporated conversational chatbot

Abstract

Almost 50% of cases of major depressive disorder go undiagnosed. In this paper, we propose a passive diagnostic system that combines the areas of clinical psychology, machine learning and conversational dialogue systems. We have trained a dialogue system, powered by sequence-to-sequence neural networks that can have a real-time conversation with individuals. In tandem, we have developed specific machine learning classifiers that monitor the conversation and predict the presence or absence of certain crucial depression symptoms. This would facilitate real-time instant crisis support for those suffering from depression. Our evaluation metrics have suggested this could be a positive future direction of research in both developing more human like chatbots and identifying depression in written text. We hope this work may additionally have practical implications in the area of crisis support services for mental health organisations.