What to Discuss?—A Blueprint Topic Model for Health Coaching Dialogues With Conversational Agents

Tessa Beinema*, Harm op den Akker, Hermie J. Hermens, Lex van Velsen

*Corresponding author for this work

Research output: Contribution to journalArticleAcademicpeer-review

6 Citations (Scopus)
118 Downloads (Pure)

Abstract

Conversational agents (CAs) are often included as virtual coaches in eHealth applications. Tailoring conversations with these coaches to the individual user can increase the effectiveness of the coaching. An improvement for this tailoring process could be to (automatically) tailor the conversation at the topic level. In this article, we describe the design and evaluation of a blueprint topic model for use in the implementation of such topic selection.

First, we constructed a topic model by extracting actions from the literature that a CA as coach could perform. We divided these actions in groups and labeled them with topics. We included literature from the behavioral psychology, relational agents and persuasive technology domains. Second, we evaluated this topic model through an online closed card sort study with health coaching experts.

The constructed topic model contains 30 topics and 115 actions. Overall, the sorting of actions into topics was validated by the 11 experts participating in the card sort. Cards with actions that were sorted incorrectly mostly missed an immediacy indicator in their description (e.g., the difference between “you could plan regular walks” as opposed to “let’s plan a walk”) and/or were based on behavior change techniques that were difficult to translate to a conversation.

The blueprint topic model presented in this article is an important step towards more intelligent virtual coaches. Future research should focus on the implementation of automatic topic selection. Furthermore, tailoring of coaching dialogues with CAs in multiple steps could be further investigated, for example, from the technical or user interaction perspective.
Original languageEnglish
Pages (from-to)164-182
Number of pages19
JournalInternational journal of human-computer interaction
Volume39
Issue number1
Early online date9 May 2022
DOIs
Publication statusPublished - 2 Jan 2023

Keywords

  • Conversational agents
  • Health behaviour change
  • Coaching
  • Dialogues
  • Topics
  • Tailoring
  • Personalisation
  • UT-Hybrid-D

Fingerprint

Dive into the research topics of 'What to Discuss?—A Blueprint Topic Model for Health Coaching Dialogues With Conversational Agents'. Together they form a unique fingerprint.

Cite this