Font Size: a A A

Robust Collaborative Recommendation Based On Shilling Attack Detection And Bayesian Probabilistic Matrix Factorization

Posted on:2016-08-04Degree:MasterType:Thesis
Country:ChinaCandidate:C J DongFull Text:PDF
GTID:2308330479451049Subject:Computer technology
Abstract/Summary:PDF Full Text Request
Due to the sparsity of the ratings and the existence of malicious users, the existing recommender systems generally have the following problems: Collaborative recommender systems are vulnerable to shilling attacks, in which malicious users can bias the systems’ output by injecting a large number of fake profiles; the robustness of the recommendation algorithms is relatively poor when facing shilling attacks and noise; some robust recommendation algorithms suffer from low robustness or improve the robustness of the system by sacrifying the accuracy in the presence of shilling attacks. To solve these problems, we propose a robust collaborative recommendation algorithm based on shilling attack detection and Bayesian probabilistic matrix factorization, in order to improve the prediction accuracy and the robustness. We mainly focus on the following aspects.Firstly, we propose a method called EPCA for identifying attack profiles in order to reduce the impact of shilling attacks. We get the set of suspicious users based on the rank of PCA. Then we can find the attacked item on the analysis of mean shift value. Then we remove the genuine users from the suspicious users set. Our algorithm can identify the malicious users and attacked item precisely.Secondly, we devise a robust recommendation algorithm based on Bayesian probabilistic matrix factorization, which incorporates the user rating matrix, item attributes. The incorporation of item attributes can reduce the effection of rating matrix on the item feature matrix, the can guarantee the robustness of the system.Then, we get the robust algorithm based on shilling attack detection and Bayesian probabilistic matrix factorization by fusing the result of EPCA algorithm and the robust recommendation algorithm mentioned above.Finally, we take experiment on Movie Lens data sets to test the robustness and the accuracy of the robust recommendation algorithm based on Bayesian probabilistic matrix factorization and shilling attack. Experiments show that the proposed algorithm can effectively improve the robustness and recommendation accuracy.
Keywords/Search Tags:robust collaborative recommendation, shilling attack, Bayesian probabilistic matrix factorization, Principal component analysis
PDF Full Text Request
Related items