Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn about popularity bias in recommender systems through this 24-minute conference talk that examines one of the most significant challenges facing modern recommendation algorithms. Explore how recommender systems often disproportionately favor popular items over niche content, creating potential limitations for both users seeking diverse recommendations and content providers trying to gain visibility. Understand the multifaceted origins of popularity bias, including inherent data imbalances where popular items naturally receive more interactions, algorithmic tendencies that amplify these patterns, and self-reinforcing feedback loops that perpetuate bias in dynamic recommendation processes. Discover various mechanisms designed to control and mitigate this bias, from pre-processing techniques that adjust training data to in-processing model modifications that account for popularity during learning, and post-processing approaches that rerank recommendations. Examine advanced strategies including dynamic debiasing methods that adapt to changing popularity patterns and innovative approaches like False Positive Correction that address specific aspects of bias propagation. Challenge the conventional assumption that popularity bias is universally detrimental by exploring scenarios where recommending popular items can actually benefit users and examining the complex trade-offs between bias mitigation efforts and overall user experience optimization in real-world recommendation systems.
