A recommendation algorithm for sparse data via implicit relationship

Abstract

The sparseness of user-to-item rating is one of the main factors affecting the quality of a recommender system. To address the sparsity problem, several recommendation techniques have been proposed. They add auxiliary information to the recommender system to improve the accuracy of rating prediction. We observe that there are more implicit relationships besides the explicit connection between users and the items to be predicted. If these rich implicit relations are introduced into the recommendation model, the sparsity of the original data can be greatly reduced. Therefore, we propose a new recommendation algorithm via implicit relationship, which is applied to the classic movie recommendation problem. This model extracts the implicit relationship from the movie's cast data to make movie recommendations, so as to better understand the movie. Different from the existing context-based methods, our method extracts more diverse preference information of users and integrates the implicit relationship between users and other entities. This paper takes a step forward in this direction, and studies how to use rich implicit relations to alleviate the sparsity of recommendation and further improve the recommendation methods. We conduct a series of experiments on real datasets which show that our method sianificantly outperforms state-of-the-art methods.