Popularity bias is a phenomenon present in recommender systems where the model focuses more on popular items, since they are the items with many more interactions, and so it will recommend them more than unpopular items, causing the recommendation to be less personalized, and with a lower serendipity.
Tt also causes the terrible Matthew effect, where popular items become even more popular by being frequently recommended1.
The phenomenon of popularity bias is caused by the recommender system general training paradigm, where popular items more frequently achieve a lower loss if recommended more1.
tags: recommender-systems