A new collaborative filtering algorithm based on modified matrix factorization
Hanmin Ye, Qiuling Zhang, Xue Bai · 2017
Recommendation algorithm based on matrix factorization has a global presented objective function by optimization technology, in which singular value decomposition is a typical representative, but there are still some bottleneck restricting the further development of problems such as high-dimensional sparse problem. Concerning that problem, we propose an alternating least square based on singular value decomposition algorithm. Firstly, we fill the user-item rating matrix with each item's mean score. Secondly we use singular value decomposition to identify the best potential factor dimension and initialize the potential factor matrix of users and items. Thirdly we use alternating least squares to get the final potential factor matrix of users and items. Finally, we use the final potential factor matrix of users and items to recommend. The results on the Movielens datasets show that the proposed algorithm can effectively improve the recommendation accuracy so as to ease the high-dimensional data sparsity.