Advancing Recommender Systems: Embracing Personal Agent-Mediated Approaches

Recommender systems have traditionally been viewed as ranking mechanisms where platforms monitor user behavior to curate item suggestions. However, this perspective overlooks the significant control that platforms wield over various aspects of the recommendation process, including access to candidates, the boundaries of evidence, and the rationale behind suggestions. The emerging challenge in this domain is not only about understanding user preferences but also about managing how evidence is gathered and shared. We propose a novel approach termed Personal Agent-Mediated Recommendation (PAMR). This framework positions a personal agent as the user's representative, facilitating the discovery, filtering, and aggregation of recommendation evidence from diverse sources. The transition here is crucial; it shifts the focus from platform-driven ranking to user-centric evidence management. In this position paper, we delineate the PAMR paradigm, outline its defining characteristics, and highlight essential mediation decisions. Furthermore, we introduce a new evaluation framework centered on mediation. Initial findings from a study on restaurant recommendations via Yelp indicate that effective source selection and controlled information sharing can significantly enhance user experience, offering improved traceability and utility.