Matryoshka Attack: Research On An Attack Method Of Recommender System Based On Adversarial Learning And Optimization Solution

Currently, recommendation systems have been widely used in various fields, especially profitable ones. However, the research on the attack method of the recommendation system is not very common, and the analysis of the reaction of the system after the attack is also rare. At the same time, most of the existing research focuses on discussing a single theoretical level of attack without taking actual conditions into consideration. To this end, we propose an evaluation model based on minimum attack resource consumption, and innovatively propose a data injection attack method based on adversarial learning. In this article, we first introduced the principle of the recommendation system, explained the effectiveness of the data injection attack on the recommendation system, and then used the new "Russian doll attack" method to strengthen the common data injection attack and apply it to three realities. To verify the effectiveness and portability of our attack. In addition, we were also pleasantly surprised to find that the Russian doll attack is particularly destructive to the popular ITEMAE recommendation system.