Dialogue System for Early Mental Illness Detection: Towards a Digital Twin Solution
Mental health disorders rates have increased in recent years, especially after the COVID-19 outbreak. Several factors contribute to this, including stigma towards mental health problems, the misconception that treatment of such conditions is only available to the ones that can afford such a luxury and hence not for the majority. In this project, we develop a dialogue system that analyses the mental status of the user to give personalized feedback based on the severity of the mental health problem. We propose a framework concept of Digital Twin for human health, where our findings contribute towards themental health component. We investigate the incorporation and effectiveness of recent advancements in Natural Language Processing, such as conversational AI and social media sentiment analysis in mental health domain. Due to high social acceptability and easiness, a chatbot framework was developed to provide mental health assessment. The conversational flow was designed in collaboration with clinical psychiatrists. For detection of various severity of mental health problem, we build a customized classification model that leverages pre-trained BERT models that is fine-tuned on E-DAIC dataset. Mental health status is further investigated by obtaining sentiment scores from social media platforms (Instagram and Twitter). Our method has the potential in detecting signs of mental health problem based on their severity levels with 69% accuracy and demonstrates high acceptability and usability (84.75%).
A. Abilkaiyrkyzy, "Dialogue System for Early Mental Illness Detection: Towards a Digital Twin Solution", M.S. Thesis, MBZUAI, Abu Dhabi, UAE, 2022.