The Study Of Robust Recommendation Method For Collaborative Filtering Recommendation

Collaborative filtering recommendation,as one of the most widely used technologies in the field of e-commerce,is an important technical means to solve the "information overload".However,in the face of shilling attack,the existing collaborative recommendation algorithms are not robust enough,and the recommendation process of the system is prone to be disturbed,which seriously affects the recommendation results of the system.To solve this problem,based on the basic matrix factorization technology,this paper adopts supervised and unsupervised machine learning algorithms to design a series of collaborative filtering recommendation algorithms with high robustness and less loss of accuracy from the perspective of attack detection.Firstly,a matrix factorization recommendation algorithm based on particle swarm optimization is proposed.In order to improve the robust estimation performance of the item feature matrix and user feature matrix,we adopt particle swarm optimization technology to initialize them according to the user rating data and the basic matrix factorization method.The obtained item feature matrix and user feature matrix can improve the recommendation accuracy.Compared with the existing algorithms on the Movielens dataset,the proposed algorithm is proved to be effective in terms of recommendation accuracy.Secondly,a robust recommendation algorithm which combines hierarchical clustering and MFRA?PSO is proposed.Based on the user rating data,we mine the individual differences between the real users and the attack users.And the attack profile detection algorithm is designed by combining with the hierarchical clustering algorithm.Then,the user profiles are divided into two categories by using the detection algorithm.We propose the defination of average inner-class distance which can identify the class of attack profile.We combine the identification results of attack profiles with the matrix factorization recommendation algorithm based on particle swarm optimization to construct a robust recommendation algorithm.Compared with the existing algorithms on the Movielens dataset,the proposed algorithm is proved to be effective from two aspects of robustness and recommendation accuracy.Finally,a robust recommendation algorithm combining random forest and target item identification is proposed.Based on the user rating data,we give the defination of item popularity and utilize Chi-square statistics to extract the effective features that can distinguish true users and attack users.Based on the extracted features,the random forest classifier is trained to detect the attack users in the first stage,and then the class of attack profile is further detected by identifying the target item in the second stage.Based on the attack profile detection results and the matrix factorization recommendation algorithm based on particle swarm optimization,a robust recommendation algorithm is designed.Compared with theexisting algorithms on the Movielens dataset,the effectiveness of the proposed algorithm is verified.