Collaborative Filtering for Recommending Movies

Abstract

There is a significant amount of ongoing research in the collaborative filtering field, with much of the research focusing on how to most accurately give item predictions to a user, based on ratings given by other users with similar rating patterns. The objective of this project is to build movie rating prediction models with a simple and intuitive representation, based on previous work within the area. Important factors are the investigation of the predictive power of these models, and the research on how the use of content information can improve accuracy when the available data is sparse. We show that latent class models provide an expressive, but yet simple way to represent the movie rating scenario, and that the models have great potential when it comes to predictive accuracy. We conclude that the inclusion of additional content features into the models can help improve the accuracy when there is little data available.