Researches On Shilling Attacks Detection Algorithm In Collaborative Filtering Recommender Systems

With the development of the mobile Internet,there is a dramatic surge in data people receive in various ways,which makes information overload a headache for contemporary people.To get this settled and improve the efficiency of the way people obtain information,this paper deploys recommendation systems in various network platforms.Currently,the most popular system is rooted on Collaborative Filtering(CF)technology.The collaborative filtering technology gleans the preferences and habits of the target user through statistics of user historical scores,and foresees the preferences of the target user judging from the selection algorithm.Even so,due to its nature of openness,CF technology is extremely vulnerable to Shilling Attacks.By infiltrating false users into the system,Attackers manipulates the prediction results of the recommendation system,which is the primary vulnerability of the recommendation system.To fix this loophole,this paper puts premium on an insightful study on the detection of shilling attacks in the centralized recommendation system and the privacy protection and detection of shilling attacks in the distributed recommendation system.The mechanism as follows:(1)Design a shilling attack detection algorithm in light of the dispersion of user ratings.By comparing the distribution of the dispersion of real users and false user ratings,three characteristics of the extreme score ratio of user ratings,the range of de-extreme ratings,and the standard deviation of user ratings are initiated as the measurement standards for the dispersion of user ratings,which hence is taken as ID3.The classification attributes of the decision tree are created and provisioned,the informative gain rate of each attribute is calculated and the attribute with the biggest informative gain is selected as the root node,and real users and fake users are taken apart in the generated decision tree,so as to achieve the mechanism of the detection of shilling attacks.The experimental results suggests that the shilling worthy attack detection algorithm based on the dispersion of user ratings has a excellent detection effect on shilling worthy attacks and the algorithm is robust.(2)Design a privacy-sensitive distributed algorithm for shilling attack detection.First,randomize the unrated items the users submit before data exchange performs,and then start the exchange data procedure according to the ADD model.Secondly,according to the features of distributed collaborative filtering,the classification attributes of the detection for specific attack models are boosted,and the hemimorphic encryption technology is used for protection during the exchange of specific attributes.Finally,all user classification attributes are collected and combined with the KNN algorithm to identify real users and fake users,so as to realize the detection of shilling attacks while securing privacy.The experimental results show that the privacy-sensitive distributed recommendation system shilling attack detection algorithm can achieve the detection of shilling attack users while securing privacy.The paper here discusses and dedicates to the problem of shilling attack detection in centralized recommendation systems and distributed recommendation systems.The objective is: to detect the referral attack in the recommendation system,minimize the impact of the referral attack on the recommendation system,and enhance accuracy of the prediction for user's preferences.With regards to the privacy leakage during data exchange in the distributed recommendation system,it secures user privacy when detecting shilling attacks.The experimental results show that the algorithm designed in this paper plays a significant role in terms of detecting shilling attacks.