Research And Implementation Of Support Attack Detection System Based On Central Features Of Graph Nodes

The emergence and rapid development of the Internet has brought massive amounts of information to people.However,as the explosion of information increases,users can obtain more information but become more and more noisy,and the recommendation system is considered to be a good solution to this problem.Unlike search engines,which actively obtain information,the recommendation system performs personalized calculations on user behavioral interests,finds user needs,and guides users.The earliest and most successful technology used in the recommendation system is the collaborative filtering recommendation technology.It uses the nearest neighbor method to use the user's historical information to calculate the distance between users,and then predicts the target user by the nearest neighbor's preference for the product to determine whether to recommend the user.This recommendation technology is very effective and can deal with unstructured objects such as video and music.However,this method also allows the attacker to find a loophole.The attacker injects false information into the recommendation system by simulating normal user behavior.The effectiveness of the recommendation result has been greatly affected.Such attacks are called “attack attacks”.The existing shilling attack detection algorithms mainly start with scoring features such as the PCA-SAD algorithm,the Semi-SAD algorithm,or obtain statistical information to obtain features such as the Degree-SAD algorithm.This article proposes a new idea of attack detection.We see the recommendation system as a complex network diagram structure.The user in this figure structure represents the user node and the project represents the project node.The user's rating of the item represents the presence of edges between the nodes.We master the graph structure features by studying the importance of nodes(node-centric features).The anomalous distribution of node centroid values is likely to represent the user's abnormal behavior.Through a large number of experimental comparisons,we find that the central characteristics of the nodes have a good effect on the detection of caret attacks.The central features combined with the classification algorithm have led to the detection of the caret attacks proposed in this paper named Cdn-SAD algorithm(Central distribution of nodes-shilling attack detection).This article starts from the following aspects:(1)Constructing the shilling attack model under the recommendation system,and analyzing the existing recommendation technology principles.At the same time,it summarizes the various central features of nodes in the graph structure.(2)The central distribution of project nodes is proposed,and the central characteristics of user nodes are proposed.Combining the features with the Naive Bayesian classifier based on EM algorithm,a Cdn-SAD algorithm based on the central feature of graph nodes is proposed.A variety of attack models were implemented and the attack was injected on the MovieLens100 K data set.Experiments were compared with algorithms such as Degree-SAD,PCA-SAD,and Semi-SAD.At the same time,experiments were performed on the fully labeled dataset Amazon.The F1 value increased by 26.36%,39.49%,and 34.97%.(3)This system is based on the MVC framework and is divided into two major modules according to the function of the system,injecting attack modules and attack detection modules.The attack injection module mainly implements common attack injection methods,such as random attack,average attack,and popular attack.The attack detection module mainly implements data set preprocessing,feature value acquisition,classifier construction,and attack detection.The implementation of the shilling attack detection algorithm completes the module design,and finally tests the operating effect of the system.