Singular Value Decomposition Algorithm Combined with SDAE for Improving the Accuracy of Movie Recommended System

Abstract

The sparseness of the rating data seriously affects the recommendation accuracy of the collaborative filtering algorithm. Since the current recommendation algorithm does not fully consider the item attribute matrix information, this paper proposes a Singular Value Decomposition (SVD) algorithm combined with Stacked Denoising Autoencoder (SDAE). The S-SVD algorithm first reduces the original user-item rating matrix by SVD. Then, we use SDAE for feature learning to calculate the similarity between the rating-based and attribute-based items on the item matrix. So, we can calculate their item the similarity between any two items which can be films, books and so on. Finally, we get the nearest neighbor set of unrated items and predict the rating to generate recommendations. Experiments on the real datasets show the S-SVD algorithm can overcome the data sparsity problem to a large extent and is superior to other traditional recommendation algorithms in performance.