Group Attack Detection Algorithm Based On Target Item Identification In Recommendation System

The popularity of the Internet has brought convenience to people's lives and travel.But,there is a lot of information in the Internet,people need spend a lot of time to filter information in order to get the information that they want.This phenomenon is called information overload.The recommendation system came into being.However,due to the openness of the recommendation system,many malicious users inject a large amount of attack into the recommendation system in order to achieve their own purposes,which causes the recommendation result to change and seriously damages the stability of the recommendation system.Many experts have proposed some solutions to this problem.However,as the attack mode gradually shifts to group attack mode,the existing methods cannot detect the group attack well.Aiming at this problem,this thesis conducts an in-depth study on group attack detection,and proposes two group attack detection algorithms based on target item recognition.Firstly,the target item rating are analyzed from the perspective of time,and use the semi-supervised algorithm to identify the target item.For each item,the item-time series is divided,and suspicious items and time interval sets are extracted.Through calculate the suspicious item characteristics and the semi-supervised classification process,the target item is identified,and then the attack group is obtained.Secondly,by analysis the user rating behavior,combined with the rating situation between users and projects,the group attack detection is studied.By extracting the characteristics of the attacking user,clustering the suspicious user set.By constructing the user rating matrix through the suspicious user set and the item,calculating the suspiciousness of the item by rating in the matrix,and ranking the suspicious degree to obtain the user suspiciousness.Then the attack group is obtained.At the end of the thesis,the results of the two algorithms proposed in this paper are compared with the existing algorithms by experimenting on Amazon,Netflix and MovieLens data sets respectively,and the effectiveness of the algorithm is verified.