Recommender system for music CDs using a graph partitioning method

Abstract

Collaborative filtering is used for the prediction of user preferences in recommender systems, such as for recommending movies, music, or articles. This method has a good effect on a company's business. E-commerce companies such as Amazon and Netflix have successfully used recommender systems to increase sales and improve customer loyalty. However, these systems generally require ratings for the movies, music, etc. It is usually difficult or expensive to obtain such ratings data comparison with transaction data. Therefore, we need a high quality recommender system that uses only historical purchasing data without ratings. This paper discusses the effectiveness of a graph-partitioning method based recommender system. In numerical computational experiments, we applied our method to the purchasing data for CDs, and compared our results with those obtained by a traditional method. This showed that our method is more practical for business. © 2009 Springer Berlin Heidelberg.