Research On Shilling Attack Detection Algorithms Based On LDA Model And Grey Theory

Collaborative filtering system to a large extent alleviate the "information overload" problem on the network,but because of its open nature,it will face vulnerability in the face of attacks,seriously jeopardizing the recommendation system.Therefore,how to effectively detect the attacking user becomes an urgent problem to be solved in the field of recommending system security.In order to control the influence of shilling attack on collaborative filtering system,many detection methods are used by domestic and foreign researchers.However,supervised detection methods are often constrained by types of attacks.However,unsupervised detection methods mostly require prior knowledge.In order to ensure its effectiveness,and in order to avoid attack detection algorithm,attacker using the new attack strategy,the existing detection algorithm proposed a great test.In response to this problem,this paper conducts an in-depth study on the detect shilling attack in collaborative filtering system by analyzing user rating behavior.First of all,by analyzing user rating behavior,this paper proposes a shilling attack detection algorithm based on user rating behavior.The detection algorithm based on the extraction of the subject of preference,quantitative analysis of genuine user and attack user rating model differences.Drawing on the idea of label dissemination,we calculate product suspicion and calculate user suspiciousness by linear weighted average.By calculating the sum of the users' suspicious differences in the sliding window,the demarcation point of the suspiciousness between the real user and the attacking user can be identified,and the size of the attacking user can be obtained so as to detect the attack user.Secondly,based on the theory of gray model and density clustering algorithm(DBSCAN),this paper presents a kind of shilling attack detection method based on gray theory and density clustering.Based on the gray theoretical model,the algorithm calculates the gray mean degree of the product mean value.Based on this,the weighted gray relational degree between users is calculated and the user distance matrix is obtained.The user distance matrix is used to conduct density clustering analysis to obtain candidate attack users.By calculating the minimum distances for each user in the attacking user andsorting,a minimum distance sequence is obtained,and the curvature of each user point in the sequence is calculated.At the same time,we select the curvature point as the demarcation point of the real user and the attack user through certain conditions to detect the attack user.Finally,the experimental results on the MovieLens1 M dataset and the NetFlix dataset show that the proposed algorithm can effectively detect many types of attacks.