Research On Shilling Attack Detection And Defense Algorithm For Collaborative Filtering Recommender Systems

With the rapid development of Internet technology, especially the popularity of mobile Internet devices in recent years, brings us into the era of rich information, but also make us lost in the amount of information. Recommender system as a kind of emerging information filtering technology should be transported and it can effectively solve the information overload problem, which applys the history of users’ rating behavior to provide personalized recommendation for human. However, collaborative filtering is facing some challenges with the open environment of the recommendation system itself. Attackers for commercial competition, artificially to the system into a lot of shilling attacks, in an attempt to make the system produce the recommended results in their favor. Therefore, how to ensure the safety of the work of t he recommender system has become an important issue. Our work from the care of attack detection and supporting attack defense two points of view: for the recommended safety problems of the system, we proposed three methods to solve this problem effectively., respectively according to different shilling attack model and recommender algorithm robustness. We comprehensively analyze the research status at home and abroad, and in depth research on the security of collaborative filtering recommendation system, the major works include:(1) in order to improve the detection effect of the average attack in the recommender system, a non negative matrix factorization based support attack detection algorithm model is constructed. Using non negative matrix factorization technique to the original user item rating matrix for feature extraction, then the clustering algorithm to extract the characteristics of clustering. Finally with the help of clustering results, with a set of normal user characteristic information and support the attack set feature information of the difference of the secondary classification, so as to improve the precision of the mean attack detection.(2) in order to improve the detection effect of random attack in the recommender system, a new detection algorithm based on density clustering is constructed. The clustering principle of density clustering algorithm is applied to the detection of random attack in the recommender system. The working principle of the algorithm is analyzed, and the method is app lied to detect the random attack. Experimental results show that the proposed algorithm can effectively detect the random attacks in the recommendation system, and has high detection efficiency.(3) in order to improve the robustness of the algorithm in the recommender system, a new model of support attack defense Algorithm Based on matrix decomposition is constructed. First, the use of attack detection technology, the detected results as a measure of whether the user is a shilling attack probability; then, in order to construct a trust weight matrix; finally, the the weight matrix to matrix factorization model and solving the model is introduced. Experimental results show that the proposed algorithm is more effective in resisting the support attack compared with other collaborative filtering algorithms.