Investigating conversational agents in healthcare: Application of a technical-oriented taxonomy

Abstract

Conversational agents (CA) are increasingly applied to realize health applications that collect patient data, provide information or even deliver health interventions. We developed a taxonomy focusing on technical characteristics of health CA with the purpose of creating a reporting guideline towards health CA and of building technical-oriented archetypes. The taxonomy comprises 18 dimensions which can be grouped into four perspectives. In this work, we wanted to find out whether the taxonomy is complete and can be applied appropriately by researcher to describe the technical characteristics of their health CA. Through a literature review, we identified 103 unique health CA for which publications have been published in 2021 and 2022. We contacted the corresponding or first authors of those papers asking for providing the information along our taxonomy for the CA described in their paper. For this purpose, our taxonomy was transformed into a questionnaire. To study applicability and understandability of the taxonomy, we also extracted the requested information from the papers using the taxonomy and compared the results to those of the participants. 95 E-Mails could be delivered. 26 persons out of 95 replied to our request resulting in a return rate of 27.3%. Results show that the majority of CA is simple in terms of CA personality; visualized as avatar or without embodiment. Systems are mainly rule-based, domain-specific and support one language. We recognized several differences between replies given by the participants and what has been extracted from the publications on the CA by us. We conclude that in order to apply the taxonomy as reporting guideline clear definitions must be given for the single characteristics. Some additional characteristics have to be added.