Accepted for/Published in: Journal of Medical Internet Research
Date Submitted: Aug 1, 2022
Date Accepted: Dec 19, 2022
Developing a Technical-oriented Taxonomy to Define Archetypes of Conversational Agents in Healthcare: Systematic Review and Cluster Analysis
ABSTRACT
Background:
The evolving of artificial intelligence (AI) and natural language processing generates new opportunities for conversational agents (CA) that communicate and interact with individuals. In the health domain, CA became popular since they allow simulating the real-life experience in a healthcare setting, which is the conversation with a doctor. However, it is still unclear which technical archetypes of health CA can be distinguished. Such technical archetypes are required among other things for harmonizing evaluation metrics or for describing the landscape of health CA.
Objective:
The objective of this work is to develop a technical-oriented taxonomy for health CA and to characterize archetypes of CA in healthcare based on their technical characteristics.
Methods:
We develop a taxonomy of technical design elements for health CA based on scientific literature, empirical data and by applying a taxonomy development framework. To demonstrate the applicability of the taxonomy we analyze the landscape of health CA of the last years based on a systematic literature review. To form technical design archetypes of health CA, we apply a k-means clustering method.
Results:
Our taxonomy comprises 18 unique dimensions belonging to 4 perspectives of technical design elements (setting, data processing, interaction and agent appearance). Each technical dimension consists of 2 to 5 characteristics. The taxonomy is validated based on 173 unique health CA that have been identified out of 1,671 initially retrieved publications. The 173 CA were clustered into 4 distinctive archetypes: 1) a text-based ad-hoc supporter, 2) a multilingual, hybrid ad-hoc supporter, 3) a hybrid, single language temporary advisor and finally, 4) an embodied temporary advisor, rule-based with hybrid input/output options.
Conclusions:
The current landscape of CA in healthcare is rule-based, often text-based and rather simple in terms of interaction and CA personality. Information related to data processing – which is part of the taxonomy – is often missing in scientific papers on health CA. We conclude there is a need for a harmonized presentation of technical details on health CA in scientific literature. Applying our taxonomy as reporting guideline might help in overcoming this limitation. The archetypes can form the basis for harmonizing evaluation procedures for each archetype. Clinical Trial: n/a
Citation
Request queued. Please wait while the file is being generated. It may take some time.