Files

No attached file found for this publication.

Details

Authors
Abstract
Recommendation systems are essential for navigating the abundance of online content and providing personalized suggestions to enhance user engagement and satisfaction. However, their reliance on personalization risks reinforcing "filter bubbles," a phenomenon where users are repeatedly exposed to content that aligns with their existing preferences. This self-reinforcing cycle can lead to reduced consumed diversity—the variety of content users actually engage with—ultimately limiting exposure to novel ideas, reducing critical thinking, and exacerbating societal challenges like polarization and misinformation. To address this, traditional approaches often focus on recommended diversity, which balances the variety of items presented to users while maintaining relevance. However, there is a critical gap between recommended diversity (what users are shown) and consumed diversity (what users actually choose). Even when exposed to diverse recommendations, users may still select content closely aligned with their preferences, perpetuating filter bubbles. To address this challenge, we propose PI-adaptDiv, a novel algorithm designed to enhance the diversity of content consumed by users rather than merely increasing the variety of recommended items. PI-adaptDiv leverages a Proportional-Integral (PI) controller to dynamically adjust the diversity levels of recommendations based on user interactions. By monitoring consumed diversity and adapting recommendations in real time, the algorithm aims to mitigate the persistence of filter bubbles while maintaining high relevance and user satisfaction. The algorithm is evaluated through offline simulations using the MIND and EB-NeRD datasets and a live online experiment on a video recommendation platform. Results demonstrate that PI-adaptDiv significantly enhances consumed diversity, particularly for users at risk of filter bubbles, while maintaining a high level of recommendation quality.
Affiliations

Citations

  • APA
  • Chicago
  • FWB

Timmers, C., & Vande Kerckhove, C. (2024). PI-adaptDiv: an adaptive algorithm to prevent and escape filter bubbles. https://hdl.handle.net/2078.5/239421