How the distribution of the number of items rated per user influences the quality of recommendations

With an ever-increasing amount of information made available via the Internet, it is getting more and more difficult to find the relevant pieces of information. Recommender systems have thus become an essential part of information technology. Although a lot of research has been devoted to this area, the factors influencing the quality of recommendations are not completely understood. This paper examines how the quality of the recommendations made by collaborative filtering recommender systems depends on the distribution of the number of items rated per user. Specifically, we show that its skewness plays an important role for the quality attained by an item-based collaborative filtering algorithm.