Consensus or conflict? Recommender systems as recursive exposure systems in the production of polarization

Recommender systems are central to debates over whether digital platforms fragment public discourse or sustain exposure to diverse information, yet empirical findings remain heterogeneous. This study examines how filter bubbles, echo chambers, and polarization are conceptualized and operationalized in recommender-system research and integrates evidence across algorithmic, behavioral, relational, and outcome levels. A systematic review of 60 studies published between 2020 and 2025, conducted and reported in accordance with PRISMA, informs the development of the Platform-Mediated Polarization Framework (PMPF). Extending the recursive logic of Reinforcing Spirals Theory to computationally adaptive recommender environments, PMPF conceptualizes recursive exposure production as a process through which behavioral responses to prior exposure may be incorporated into subsequent ranking and personalization. Evidence was more extensive and direct for recommender-mediated structuring and differentiation of informational exposure than for attitudinal or ideological polarization. Evidence for polarization was more heterogeneous, suggesting that such outcomes are contingent on behavioral, relational, informational, and contextual processes. Direct evidence of recursive feedback was limited to a subset of studies, while others examined component processes or addressed feedback conceptually. The evidence therefore does not support a uniform relationship between recommender systems and polarization. Instead, PMPF integrates these patterns within a system-level framework in which exposure conditions may be recursively produced across successive interactions, while their direction and consequences remain open to empirical investigation.