Evaluating Trust in Recommender Systems: A User Study on the Impacts of Explanations, Agency Attribution, and Product Types

Abstract

Recommender systems have emerged as powerful tools for providing personalized recommendations, often employed by online platforms to suggest products or media content based on user behavior. The design of these systems significantly influences and shapes user experience, as they guide users through vast amounts of information and help them make decisions more efficiently. In this study, we conducted an online experiment (N = 268) to test how different framings of source (human-oriented vs. machine-oriented vs. proxy-oriented vs. none) in recommendation explanations induced users’ attribution of human vs. machine agency in the recommendation process, thus impacting trusting beliefs and trusting behavioral intentions. Results revealed that subtle wording variations could lead participants to orient to different recommendation sources and then attribute human or machine agencies differently, regardless of their understanding of the true technical mechanism. Participants who attributed greater human agency to the recommendations exhibited higher confidence in making choices based on the system’s suggestions and a greater willingness to disclose personal information for continued use, mediated by the competence and integrity dimensions of trust, respectively. By evaluating both wine and vacuum recommendations, we also explored the contextual differences between hedonic and utilitarian product types in the recommendation process. The findings of this study provide implications for the trust-building and ethical communication design of recommender systems.