Research On Shilling Attack Detection For Collaborative Filtering Recommender Systems

The deep integration of modern society and digital technology has provided a powerful source of power for the interconnection of massive amounts of information,but,at the same time,it has also led to information overload for ordinary individuals.In this context,collaborative filtering recommendation systems have solved information overload problem while significantly reducing transaction costs between platforms and individuals,providing the underlying technical support for continuous innovation of business models.Unfortunately,in order to maximise malicious shilling attack user illegal profits,malicious attackers exploit the vulnerabilities of collaborative filtering recommendation systems and launch organized malicious shilling attack,seriously disrupting the commercial security of normal society.In order to effectively deter and combat malicious attackers and maintain the stability and accuracy of the output results of the collaborative filtering recommendation system,the main contents are as follows:1.In view of the limitations of some existing attack detection methods,the entry angle of rating value is relatively single and the classifier is limited,this paper proposes a double angle malicious shilling attack detection method(GIT-SAD)based on the gradient lifting decision tree.The framework of the method takes the double-angle fusion as the basic algorithm idea.Firstly,the process of users’ scoring behavior is studied,and then the two-angle algorithm is constructed by the fusion of the rating distribution matrix and time window,and the TPUS-DB algorithm is proposed.Secondly,the BCC-G algorithm is proposed based on the gradient boosting decision tree to complete the generation and integration of the base classifier set.Finally,in the malicious shilling attack detection stage,The method framework proposed in this chapter is used to detect and identify malicious shilling attackers.2.In view of the limitations of some existing shilling attack detection algorithms that do not pay enough attention to the target items and have limited detection accuracy,this paper proposes a shilling attack detection method that combines penalty factor and support vector machine(STP-SAD).This method firstly analyzes the behavior process of normal users and malicious shilling attack users when choosing the target project from the focus of the target project.Secondly,the average index of the target project is constructed by focusing on the factors,and the conversion index and its corresponding punishment factor are proposed.Finally,in the stage of malicious shilling attack detection,the punishment factor is fused with the support vector machine,and STP-SAD detection method is proposed to detect and identify malicious attack users.3.In view of the above two shilling attack detection methods,this paper respectively in the corresponding experimental part of multiple groups of experiments and results analysis.Firstly,in the self-performance experiment part,the effectiveness of the shilling attack detection performance of the method is analyzed from different aspects by continuously increasing the shilling attack intensity and controlling related parameter variables.Secondly,in the comparison experiment part,the comprehensive performance detection index is taken as the evaluation standard,a variety of benchmark attack detection algorithms are selected and shilling attack parameters in different proportions are configured.The experimental results show that the proposed methods in this paper can effectively detect and identify malicious shilling attacks.