Group Attack Detection Based On Multiple Data Sources

With the continuous development and improvement of network technology,we can buy our own products on various e-commerce platforms.When purchasing products,recommender system has gradually become a tool people rely on because of its excellent information filtering function.However,for the sake of commercial interests,some illegal businesses employ grey organizations to give false ratings to some products and give people wrong product recommendations.These grey organizations are active in various e-commerce platforms due to their interests,which poses a huge threat to the security of recommendation system.Therefore,how to detect these organizations effectively is an urgent security problem in recommendation system.At present,the detection algorithm is mainly studied in a single data source,ignoring the internal relationship of users in multiple data sources.Aiming at this problem,this paper proposes two group attack detection methods based on multiple data sources by analyzing the behavior of the same user in different data sources.Firstly,aiming at the behavior differences of multi-source users in different data sources,a group attack detection algorithm based on the behavior differences of multi-source users is proposed.Firstly,in a single data source,the user relationship graph is established by fusing the score and time information of users.Then,the network representation learning model sdne is used to automatically learn the potential characteristics of users.The clustering method is used to divide users with similar behaviors into a group,and the group detection index is used to calculate the group suspicious degree.Finally,the behavior of multi-source users in different data sources is used The difference determines the group attack user.Secondly,aiming at the limitation of the previous algorithm,which only uses the performance differences of multi-source users in different data sources and requires the prior knowledge of multi-source users,a group attack detection algorithm based on multi-source user feature fusion is proposed.The algorithm discusses the implicit social relationship between users in the recommendation system,and uses the generated adversary network and a small amount of multi-source user information to learn the mapping relationship between different data sources.Finally,it uses the learned mapping relationship to fuse the multi-source features of users to detect attack groups.Finally,through the detection results of multi-source synthetic datasets generated on the basis of Netflix and movielens respectively,it shows that the algorithm proposed in this paper can further optimize the detection results on a single data source,and the research direction of this paper has certain feasibility and the possibility of future development.