Getting Started in the Correct Stage: Designing Conversational Strategies for Proactive On-boarding in a Conversational Agent
Authors
Tonchev, Antonio ; Andersen, Thomas Svoldgaard Thomhav
Term
4. term
Education
Publication year
2024
Submitted on
2024-05-28
Pages
13
Abstract
This study examines how different conversational strategies in a conversational agent (CA) affect retention and trust in a physical activity behavior change program. We compared three strategies: Control (standard), Empathetic (supportive tone), and Extended Conversation (longer exchanges). The goal was to support onboarding and assess where participants were in their change process (stage assessment). Thirty participants took part in a 5-day study. We collected data on how accurately the CA assessed participants’ stage and on trust scores, and conducted post-interaction interviews. Using statistical tests (including Chi-squared) and qualitative analysis, we found nuanced results. There were no significant differences in perceived effectiveness among the strategies. Interviews highlighted improvements that matter for engagement: clearer prompts, more personalized interactions, and a balanced level of human-likeness in the CA’s responses. Despite low retention and technical issues, the study points to the potential of CA-based interventions, provided user preferences are addressed and the CA’s design is refined. Going forward, user-centered design and further research will be key to achieving positive health outcomes.
Denne undersøgelse ser på, hvordan forskellige samtalestrategier i en conversational agent (CA) påvirker fastholdelse og tillid i en indsats for at ændre fysisk aktivitetsadfærd. Vi sammenlignede tre strategier: Control (standard), Empathetic (medfølende tone) og Extended Conversation (længere samtaler). Målet var at hjælpe brugere godt i gang og vurdere, hvor de befandt sig i deres forandringsproces (stage assessment). I alt deltog 30 personer i et 5-dages forløb. Vi indsamlede data om, hvor præcist CA’en vurderede deltagernes stadie, samt deres tillidsscorer, og vi gennemførte interviews efter interaktionen. Statistiske tests (bl.a. chi2-test) og en kvalitativ analyse gav nuancerede resultater. Der var ingen signifikante forskelle i oplevet effektivitet mellem strategierne. Interviews pegede på vigtige forbedringer for engagement: tydeligere prompts, mere personalisering og en afbalanceret grad af menneskelighed i CA’ens svar. På trods af lav fastholdelse og tekniske problemer peger studiet på potentialet i CA-baserede adfærdsinterventioner, hvis brugerpræferencer adresseres, og designet forfines. Fremadrettet er brugercentreret design og yderligere forskning afgørende for at realisere positive sundhedsresultater.
[This apstract has been rewritten with the help of AI based on the project's original abstract]
Keywords