Toward Automated Evaluation of Empathetic Responses in Virtual Human Interaction Systems for Mental Health Scenarios

This paper investigates the process of automating the evaluation of empathetic response levels in virtual human interaction systems implementing mental health scenarios. Two suicidal virtual patients were developed to collect clinician participants' empathetic responses. Before collecting clinicians' responses, we tested the virtual human interaction with healthcare trainees. Trainees' empathetic responses were evaluated by experts to use the ECCS scale based on the ECCS level (Empathetic Communication Coding System). We trained classifiers using trainees' empathetic responses with experts' coded empathy levels as the training dataset. Clinician participants' empathetic responses to virtual patients were evaluated by experts and the classifiers. The performance of the classifiers was evaluated using the experts' coded level of clinicians' empathetic responses as a test dataset.This work demonstrates the applicability of using virtual agents techniques to identify empathy levels of clinicians' responses automatically. This work shows the potential of using virtual human interaction to train clinicians' skills to show empathy. Corresponding feedback could be provided to clinicians based on the evaluation results. We hope this study motivates more research in using intelligent virtual agents in personal skills training in education.