Group Attack Detection Method Based On Group Behavior Characteristics Analysis In Recommender System

As a product of the current information age,recommendation system helps people solve the problem of information overload when there is no exact demand or large amount of data,and provides high-quality recommendation for users to access portable.However,due to the fragility of the recommendation system and the complexity of network information data,the recommendation system is vulnerable to be attacked,especially group attacks,which has a great impact on the recommendation system.This is because attack groups can effectively change the personalized recommendation list provided by the recommendation system for users in a relatively short time,and thus affect the credibility of the whole recommendation system.Therefore,how to effectively detect group shilling attacks in the collaborative filtering recommendation system is an important problem to be solved.Based on the analysis of group attack characteristics,this paper conducts an in-depth study on the group attack detection of collaborative filtering recommendation system.Firstly,in order to solve the problem that the traditional methods can not effectively distinguish the attack users and normal users under the target items,this paper proposes a group attack detection algorithm based on target item analysis.By analyzing the rating data of items at first,this paper proposes two new target item identification features and uses the K-means clustering algorithm to identify suspicious target items.Then,the rating time series of each target item is divided into candidate groups according to the time interval.Finally,for each candidate group,define a new method of group suspicious degree calculation,and use the clustering algorithm to detect attack groups.Secondly,in view of the problem that the existing homogeneous network can not fully represent the user behavior,this paper proposes a group attack detection method based on user embedding.At first,by analyzing the user's rating behavior,construct a user-item heterogeneous interaction graph,and use the Struct2 Vec model to generate user feature vectors.Then,by analyzing the rating of the item,the suspicious target items are obtained.Then,for each user under the suspicious target items,the user's suspicious degree is calculated in the vector space,and the user is clustered by K-means algorithm.At last,the attack group is obtained.Finally,experiments are carried out on the Amazon dataset and the Netflix dataset respectively,and compared with the existing classical detection algorithms to verify the effectiveness of the proposed method in this article.