High Dimensional Explicit Feature Biased Matrix Factorization Recommendation

Abstract

Collaborative Filtering method using latent factor model is one of the most popular approaches in personal recommending system. It is famous for its good performance by using only user-item rating matrix. The latent progress intelligently factorizes users' preference on different items through the rating matrix. However, the factorization progress is completely implicit. Thus, it is difficult to integrate new observed features, and it becomes more complicated when one feature has multiple values. In this paper, we propose a new algorithm based on Matrix Factorization to model explicit features besides rating values by adding high dimensional factors, which makes the factorized presentation explainable. The algorithm is generally applicable for such discrete features as type, genres, age and so on. Experimental results show that our approach outperforms the state-of-the-art methods using latent factor model.