What is popularity bias in recommendation systems?

Popularity bias is the tendency of recommendation systems to over-recommend already popular items.

Popularity bias occurs because training data is dominated by interactions with popular items, which creates a feedback loop: popular items get more exposure, leading to more interactions, further reinforcing their popularity. This reduces diversity and hurts long-tail item discovery. It is especially strong in implicit feedback systems. Modern approaches mitigate it using reweighting, debiasing losses, exploration strategies, or causal inference techniques.