Empowering Users with Federated Recommendation Systems

Traditional recommendation systems often rely on centralized data collection, which can compromise user privacy and limit individual control over preferences. In contrast, federated learning enables recommendations to be generated without transferring personal data off-device. This study introduces a live federated recommender system that empowers users to dictate their recommendation preferences while maintaining data privacy. Over a 53-day period, 22 participants engaged with a catalog of 8,807 items, toggling between personalized and diversity-focused recommendations. The findings reveal that when given the choice, users favor personalized suggestions, achieving a click-through rate of 65.37% compared to 62.07% for diverse options. Participants also demonstrated a high level of engagement with the system, reflected in a satisfaction score of 3.93 out of 5 and 248 adjustments made to their settings. Furthermore, users gained insights into how their interactions influenced the recommendations they received, thanks to immediate feedback mechanisms. This research illustrates that it is feasible to merge user autonomy, privacy, and effective personalized recommendations within a functional framework.