Research On Group Attack Detection Algorithms Based On User Neighbor Model And Hierarchical Clustering

The recommendation system generates characteristics of recommended information by analyzing user preference data,so that some attack users or groups perform false scoring on the target product for their own economic interests,and interfere with the recommendation result of the recommendation system.In order to eliminate the impact of attack users on the recommendation system,domestic and foreign scholars analyze the behavior of attack users and propose some methods to identify attack users.However,in order to achieve the purpose of the attack quickly and effectively,the attack user forms a group attack through joint,and the group attack has strong concealment and is more threatening to the recommendation system.Some existing methods for detecting attack users are not able to identify group attacks well.In response to this problem,this paper analyzes the group attack behavior patterns and proposes two group attack detection algorithms.Firstly,from the perspective of user rating behavior,the difference between the ratings of real users and attack users is analyzed.A group attack detection algorithm based on user neighbor model is proposed.In the set time window,the user rating's deviation is used to measure the similarity of the behavior of the two users.The modular metric is combined with hierarchical clustering method to realize user clustering,and the clustering results are divided into candidate groups,according to the relevant group characteristic indicators.the attack group can be more obviously found,to improve the security of the recommendation system.Secondly,from the perspective of user rating time and user rating,the user behavior data is analyzed,and a group attack detection algorithm based on improved hierarchical clustering is designed.By analyzing the user's rating time characteristics and user's rating characteristics,the algorithm calculates the user's rating time relevance and the user's rating relevance,and obtains the user's relevance.By analyzing the user's relevance,to construct the user's neighbor relation graph,and the internal closely-connected base cluster is determined.The improved hierarchical clustering method is used to obtain the candidate group.And use the relevant group detection indicators to calculate the group suspiciousness to obtain the attack group.Finally,the above two group attack detection algorithms are evaluated on Amazon and Yelp datasets,and compared with the existing group attack detection algorithms to verify the effectiveness of the algorithm.