|
Abstract:
Rating prediction is an important application, and a popular research topic in collaborative filtering. However, both the validity of learning algorithms, and the validity of standard testing procedures rest on the assumption that missing ratings are missing at random. In this talk I will present the results of a user study and survey conducted at Yahoo! Research. The study is aimed at collecting rating data to properly validate models when user-supplied ratings may not be missing at random. I will present an analysis of the rating data collected, and an analysis of a user survey on rating behaviour. I will present experimental results showing that incorporating an explicit model of the missing data mechanism can lead to significant improvements in prediction performance. Finally, I will discuss current modeling extensions based on Dirichlet Process mixtures, and Conditional Restricted Boltzmann Machines.