Robust Recommendation Algorithm Based On Double Filtering Attack Detection

The recommendation system open to recommendation system platform brings great challenges,such as some premeditated organization targeted attacks,through to the recommendation system input for the attacker to get favorable rating data recommended by deviation results,meet their own interests and disrupt the recommendation system.Users use different strategies to attack the organization to attack different models,these attacks usually contain the target project model and some filling project,generally normal users will only buy their own goods evaluation according to their love,but not malicious evaluation.There are two ways to interference shielding attack profile on the recommendation results: one is to improve the anti attack capability of self recommendation system,improve its robustness in Recommendation Algorithm Based on;the other is through a profile injection attack detection and recognition and shield attack profile to improve the robustness of recommendation.In order to reduce the robustness and accuracy of the proposed model,this paper proposes a robust recommendation algorithm based on two interception attack detection and matrix decomposition.First of all,according to the characteristics of users and normal users attack profile high correlation low correlation,using principal component analysis of covariance calculation,the user marks low covariance data into suspected attack users,marking scope should be appropriate to expand the size of the attack and interception range,from the division that the suspected attack sequence and normal sequence of user user.Secondly,due to a very small number of attacks into the normal user sequence inside the user,so in the first interception of suspected users to find the normal user and filter.According to the normal user and attack user rating habits are different,the rating deviation is calculated,according to the appropriate proportion of two intercept filter out suspected attack users and normal users,and users to determine the suspected attack attack users.Finally,the sparse data matrix decomposition,using stochastic gradient descent learning,in the learning process,those convicted of interference shielding against users,so that it does not participate in the recommendation,to ensure the accuracy of the recommendation system,recommendation system and improve the anti attack capability,the robustness is improved significantly.