Assessing the value of unrated items in collaborative filtering

Abstract

In collaborative filtering systems, a common technique is default voting. Unknown ratings are filled with a default value to alleviate the sparsity of rating databases. We show that the choice of that default value represents an assumption about the underlying prediction algorithm and dataset. In this paper, we empirically analyze the effect of a varying default value of unrated items on various memory-based collaborative rating prediction algorithms on different rating corpora, in order to understand the assumptions these algorithms make about the rating database and to recommend default values for them.