Based On Synergistic Effect Of Shilling Attack In The Study Of Shilling Attack Detection Algorithms

The advent of the information age, it broaden our horizons and makes our life more convenient. At the same time, the problem is that the information overload brought trouble to the people who want to quickly find the information but need to pay the cost of more and more high. Search engine and collaborative filtering technology are two main means of solving this problem at present. It is worth mentioning that the collaborative filtering technology in terms of customization to the user with strong appeal. With the development of e-commerce, collaborative filtering technology is integrated into them and become the core of the recommendation system, the emergence of the recommendation system greatly enhance the user’s shopping experience, to listen to music. However, due to the open nature of the recommendation system, it is easy to suffer from the attacks of malicious users, the attacks directly affect the user’s experience, and indirectly affect the survival of e-commerce.At first, our paper analyzes the current research on this problem, the main solution technology at home and abroad, at the same time we discuss the attack model and the characteristics of the attack and other features are described in detail. For supporting attacks through the synergy of recommender system this question, in reference to other scholars support attack detection method, and summarizes the relevant scholars have put forward the corresponding improvement strategies, and the algorithms that are confined to the scene. This paper mainly focuses on the characteristics of the attack and the way to think and research. Identification of single user profile support attack detection algorithm would make mistakes.experts users also tend to be incorrectly labeled in the recommendation system.Firstly, based on the theory of signal denoising, this paper puts forward the improved principal component analysis method, the algorithm can effectively remove the dependence on the prior knowledge about the attack strength, and the accuracy rate has better performance, the algorithm has a better practical application value.Then, for a variety of hybrid attack type, the previously proposed algorithm effect is poorer. Thinking for this problem, this paper puts forward the combination model and the information entropy of attack detection algorithm, using the topic model to get the distribution of theme of the user, the attacked user profiles will focus, namely the corresponding information entropy smaller; On the contrary, the normal users usually contains multiple topics, namely the corresponding information entropy is bigger.Finally, the experiments have been carried out to verify the proposed algorithm, in front of the two algorithm in the contrast experiment on two different data sets and the results analysis. Results show that the proposed improved algorithm compared with the original algorithm, greatly improve the accuracy of prediction.