AI Podcasts

Fairness in PCA-Based Recommenders

In this episode, David Liu, an assistant research professor at Cornell, discusses how standard recommender techniques like PCA and collaborative filtering can unintentionally disadvantage minority and niche user groups. He introduces the notion of "power niche users"—highly active users with specialized tastes whose data can benefit the whole platform—and explains why traditional PCA tends to over‑focus on popular items, neglecting both niche content and new fans of popular artists. Liu presents his research on item‑weighted PCA and data upweighting as practical solutions that improve fairness without sacrificing accuracy, challenging the belief that fairness and performance are mutually exclusive. The conversation ties these theoretical advances to real‑world deployments at companies such as Meta.