Skip to content

Recommender systems

Recommender systems are algorithms that filter and rank content to suggest items of potential interest to individual users. They are central to modern online platforms—social media feeds, streaming services, e-commerce, news aggregation—and determine what content billions of users see daily.

While recommender systems improve user experience by personalizing content discovery, they also pose significant risks. They can amplify low-quality, sensational, or false information by optimizing for engagement metrics (clicks, time-on-platform) rather than accuracy or social value. They can discriminate against users or items, leak personal data, and be manipulated by adversaries to spread misinformation or artificially amplify certain voices.

Research on recommender systems in the misinformation context focuses on: (1) understanding how false claims and conspiracy theories spread through algorithmic ranking, (2) designing defenses against adversarial manipulation (fake accounts, coordinated inauthentic behavior), (3) measuring and reducing bias and unfairness in ranking, (4) building transparency and interpretability into recommendation decisions, (5) protecting user privacy while providing personalization, and (6) balancing engagement optimization with information quality.

Key papers

  • The Amplification Paradox in Recommender Systems — Agent-based model explaining why algorithmic audits find recommendation amplification of extreme content yet real user logs show recommendations don't drive extreme consumption; argues audits must model user preferences
  • A Comprehensive Survey on Trustworthy Recommender Systems — Comprehensive survey of trustworthy recommender systems across six dimensions: Safety & Robustness (defense against adversarial attacks), Non-discrimination & Fairness, Explainability, Privacy, Environmental Well-being, and Accountability & Auditability.