Users’ perceptions of empathic dialogue cues: a data-driven approach to provide tailored empathy

Understanding how and in what circumstances users respond to different verbal expressions of empathy will be important for designing Intelligent Virtual Agents able to influence the emotions and intrinsic motivations of users. We report a study to teach healthy study habits and tips involving 239 undergraduate students, 161 of which interacted with a character designed to express empathy through dialogue. We elicited participants’ personality and psychological state (depression, anxiety and stress levels) and attitudes to study. In this paper we present a detailed analysis of participants’ responses to specific empathic dialogue snippets designed to include one or more empathic cues to determine whether they found it helpful, stupid or empathic. We provide an example of how we have used data mining on our dataset to suggest which cues are most appropriate to which user types to enhance user models and improve agent decision-making regarding the expression of empathy.