Robust Collaborative Filtering Method Based On Matrix Factorization

Currently in the field of recommendation,Collaborative filtering recommendation is deeply studied,at the same time,it is also widely used in practice and has very important recommendation algorithm.But when the data set is mixed with attack data or when malicious users(such as shilling attack),these collaborative filtering recommendation algorithms now has been proposed,which expose their weaknesses.The traditional collaborative filtering algorithm based on matrix factorization is the least square method,however,because the method is inherently sensitive to outliers,the robustness of the proposed method is poor.Therefore,the security problem of recommender systems is facing great challenges.This paper has carried on the thorough research to this question.We mainly focus on the following aspects.First of all,this paper proposes a robust collaborative recommendation algorithm based on Tukey dual weighted M estimator.In order to improve the robustness of the commendation algorithm,a robust Tukey method is introduced.Compared with the traditional robust estimation algorithm based on the Huber M estimation,the accuracy and robustness are improved.The iterative least square method and the stochastic gradient descent method are introduced into the model.In order to achieve the robustness of the proposed algorithm has been enhanced.Secondly,this paper proposes a robust recommendation algorithm based on MM estimation.This algorithm is a new method to integrate two robust estimation methods.Firstly,the minimum residual error is estimated by S estimation based on the minimum residual scale,after introducing Tukey double weight function,weighted least squares estimation and stochastic gradient descent,through continuous iteration,the robust parameter estimation of the user characteristic matrix and the item characteristic matrix is completed.the robust recommendation algorithm based on MM estimation has high efficiency and robustness.Finally,the algorithm proposed in this paper is designed,and the efficiency and robustness of the algorithm are verified by experiments.The experimental data set is a set of Movielens data sets which are well known in the world.