Research On Shilling Attacks Detection Methods Based On Gaussian Mixed Model

With the wide spread of Internet in recent years,the Internet is changing people's lives step by step.Because of the massive resources in the Internet,it is crucial for people to choose effective information accurately and quickly.For example,if you search for flowers on Baidu,there will be thousands of flowers.We need to select our ideal flower species from thousands of flowers,namely "information overload." Recommendation system shows better performance in solving such problems.The collaborative filtering algorithm is used to analyze the user's historical operation behaviors,such as: likes,demands,interests,etc.and filters out effective information.However,some illegal users have injected illegal data into the recommendation system for profit,resulting in biased recommendation results.In order to ensure the authenticity results of the recommendation system,this paper starts with the generation of different Gaussian distributions from the feature datasets and combines the Gaussian mixture model to study the detection of the attack.Firstly,in view of the low universality and low detection efficiency of the existing attack detection methods,this paper proposes a method based on the relative information entropy Gaussian mixture model to detect the attack.The method first dynamically selects the most suitable features among all detection features.Finally,according to the optimal feature processing score results,a Gaussian mixture model algorithm is used to detect false users.Secondly,based on this,aiming at the problem of poor sparsity of scoring data sets and poor detection of AoP attacks,this paper proposes a method to detect the attack based on the stable matching Gaussian mixture model.The method firstly uses the stable matching algorithm to reduce the dimension of the existing dataset,separates the real users that clearly distinguish the spurious users,then processes the reduced-dimension dataset according to the relative information entropy,and selects the effective detection feature processing score results.Finally,the false user is detected by using Gaussian mixture model and AdaBoost algorithm.Finally,the two detection methods proposed in this paper are validated on the MovieLens 1M dataset.The experimental results are compared with the existing methods for detecting attack.