Adaptive emotion-aware chatbot for mental health diagnosis using recurrent reinforcement learning and transformer models

In the busy and stressful modern world, people tend to disregard mental health, still it is an important factor of overall health. The constant pressure to achieve success, the invasive nature of technology, and the constantly growing needs of the contemporary world may all be the causes of stress, anxiety, and other mental health difficulties. Despite growing awareness, mental health remains a sensitive topic, as social stigma and other factors continue to hinder open discourse. There are many standard tests like GAD-7 for anxiety, PHQ-9 for depression, PSS-10 for stress, and many more openly available on the internet, but people might miss estimate their situation while answering these questionnaires, leading to wrong or inaccurate diagnoses. This paper focuses on integrating the questions of these three standard questionnaires and creating an emotion-aware chatbot with a dynamic questionnaire. To evaluate user responses, i.e., to measure the severity of each mental disorder, a fine-tuned RoBERTa model is used. This fine-tuned transformer model will take in the user response and return a severity value for each of the three disorders on a scale of 0–100. The ROC curve method is applied to the MHP data-set to determine the threshold score for each question. Using the above two, a Recurrent Reinforcement Learning model is trained, which combines Proximal Policy Optimization (PPO) and Long Short-Term Memory (LSTM) to create the dynamic questionnaire. The Recurrent RL model will be trained to understand and evaluate the scores obtained from the user response and the history of the current session and dynamically decide which question to ask next.