Group Attack Detection Based On User Rating Deviation Degree And Time Series

As the number of people using the Internet increases,the amount of data becomes larger,so people need to spend a lot of time to filter useful information.This phenomenon is called "information overload." The collaborative filtering recommendation system has greatly alleviated the "information overload" phenomenon,but because of its own openness,it is vulnerable to attack and thus leads to inaccurate recommendation results.The attack of gray organization has certain tactical characteristics,which can change the recommendation result in a short time.Therefore,how to effectively identify group attacks has become a key issue that needs to be solved urgently.This paper starts with an in-depth study of group attack detection from the perspective of user rating detection and time series.The main contents of this paper are as follows.First,the existing supervised detection algorithms generally need to satisfy the certain prior knowledge to ensure the detection effect of the algorithm.This paper proposes a new detection method based on multi-dimensional user features,which proposes two group attack characteristics.The k-means clustering algorithm is used to generate two candidate suspicious groups according to the two group attack user characteristics,and then the intersection of the two candidate groups is taken as the suspicious user group.The user suspiciousness is calculated according to the group attack characteristics,and the suspicious order is sorted for the users in the suspicious group.Second,the traditional time series detection method uses the number of ratings of the project in a certain period of time to construct the sequence.This does not effectively filter out normal users under the target item and cannot effectively attack the attacking users.This paper proposes a time series based group attack detection method.The algorithm first needs to construct a project time series to divide the time window,then calculate the window rating percentage,calculate the rating entropy of each item,and finally combine the item score entropy.The product of the group attack is detected by the product of the value of the window rating.Finally,experiments on MovieLens100 k,Netflix and Amazon datasets verify the effectiveness of the proposed group attack detection algorithm..