| Due to the openness of the collaborative filtering algorithm in the recommendation system to obtain data,the recommendation system will be vulnerable to attacks by malicious users masquerading data,so as to achieve the purpose of improving the recommendation ranking of a project or reducing its ranking,and gain benefits from it,that is,trigger the "shilling attack" problem.The occurrence of shilling attack in the recommendation system has seriously affected the recommendation accuracy of the recommendation system.This paper studies the underattack detection and scoring prediction respectively,and proposes a robust recommendation algorithm.The main work is as follows.A shilling attack detection algorithm based on improved K-means clustering algorithm is proposed for shilling attack detection.Considering the universality of existing user features is not strong,we can use information entropy to select the feature with the greatest discrimination for different attack types and attack scales.Based on the selected feature,we redefine the formula for calculating the distance between users.Considering that the attacking users are a small part of the data set,the update rule of the cluster center is improved,which can better gather the attacking users into a class and facilitate the identification of attacking users.The experimental results show that our detection method proposed in this paper has better performance.We propose a matrix decomposition scoring prediction model that introduces user trust.In order to comprehensively analyze user scoring behavior data,we propose to divide user characteristics into three aspects,and propose a new feature in the time dimension.In these three aspects,we comprehensively calculate the user’s weight and reduce the suspicious users’ weight.Based on this,we build a user trust matrix and introduce it into the regularization matrix decomposition model.To limit the participation of suspicious users in the recommendation,so as to reduce their impact on the score prediction results.We integrate the attack detection algorithm into the matrix decomposition prediction model,and design a robust recommendation algorithm.Compared with other popular methods,the experimental results show that the algorithm proposed in this paper has a significant improvement in hit rate,offset rate and other indicators. |
