Research On Shilling Attack Detection Algorithm Based On Information Entropy And Trust Relationship

With the rapid development of computer and network technology,the amount of information received by people increases greatly,and even the problem of "information overload" appears.Personalized recommendation system through the analysis of historical data such as ratings and comments left by users,obtains user interest preferences,and provides accurate recommendations for users according to preferences,which shows good performance in solving the problem of "information overload".However,some illegal attackers will give the highest or lowest score to some products,so as to change the priority of the target project recommended by the system.This kind of attack is called shilling attack.With the continuous promotion and development of recommender system,the detection of shilling attack has attracted the attention of academia in recent years,and has become one of the research hotspots in the field of recommender system.This paper analyzes the problems of existing algorithms from two aspects of scoring similarity and social relationship,and proposes a shilling attack detection algorithm based on information entropy and social trust relationship,which improves the accuracy of shilling attack detection.The main work of this paper is as follows.(1)Unsupervised learning does not need a large number of labeled data,and does not rely on the feature index and training set.It can also detect unknown types of shilling attacks effectively.Based on the in-depth study of unsupervised shilling attack detection method based on personal rating information entropy,this paper proposes an unsupervised shilling attack detection algorithm based on information entropy similarity.After comparing the difference between the individual and the whole score,the information entropy is integrated into the similarity calculation of the individual and the whole,and the similarity is used to detect the attack.Experiments show that the algorithm can effectively detect common attacks in the absence of information about attack patterns,and has a high degree of unsupervised and universality.(2)In terms of social trust,most recommendation algorithms based on social networks only use the binary trust relationship between users' friends,and ignore the impact of indirect trust relationship.In this paper,a fusion trust model considering the trust relationship between users is proposed,which combines the indirect trust relationship with the direct trust relationship.Even if there is no direct trust relationship between users,there is at least one reachable path,which solves the sparsity problem of direct trust relationship.At the same time,the fusion trust model not only enhances the scope of the attacker's influence,but also reduces the intensity of the attacker's influence on the users who have direct trust relationship with them,so as to avoid the users who have direct trust relationship with the attacker being judged as the attacker directly.(3)In this paper,considering both the unsupervised shilling attack detection algorithm based on information entropy and the indirect trust relationship,we use the score prediction of the fusion of information entropy similarity and trust relationship in the recommendation system,and improve it,so as to obtain a shilling attack detection algorithm based on the fusion of information entropy similarity and indirect trust relationship.The algorithm evaluates the suspicious degree of the suspected attacker,selects several items with the highest suspicious degree,and finally obtains the shilling attacker.There are 12 figures,11 tables and 73 references in this paper.