Improving the Collaborative Filtering Recommender System by Using GPU

Abstract

As the expansion of Internet, the recommender system is attracting the attention of many industry engineers and researcher, especially the collaborating filtering recommender system. However, there are still some challenges. For example, the sparse feature and large scale system degrades the recommendation accuracy and efficiency. In this paper, we propose implied-similarity and filled-default-value methods to improve the denseness of the preference matrix and use GPU to parallel the process. Our experiments show that the accuracy can improve 20% and efficiency can speed up 4 times.