In most of the cases in recommender system dataset that contain the interaction between user and items, the frequency distribution is long-tail, meaning that the majority of interactions are occupied by a small number of popular items, while there is a long tails of interaction between a huge number of unpopular items1.

This causes the Popularity Bias, which is a problem that all recommender systems have and that has to be tackled in one way or another, in order to have more satisfying recommendations.


tags: recommender-systems

Footnotes

  1. Wei et al. - Model-agnostic counterfactual reasoning for eliminating popularity bias in recommender system