The Research Of Attack Detection In Recommendation System

The development of the Internet cause the problem of information overload.Personalized recommendation technology came into being in order to solve this problem.With the business in an increasingly competitive,some malicious users attack the recommendation system to affect the recommended result by using the characteristics of the system. In recent years, how to detect the recommended attack effectively has become a focus research.This thesis research the recommended attack detection especially focus on the detection of profile injection attack.Our research including the composition and structure characteristics of user profiles,several attack models(such as random attack) and several attack detection technology. On the basis of the research and analysis about the present situation,this thesis proposes a new decision approach to detect the profile injection attack.The main work of this thesis as follows:1. Studying the principle of recommendation system and collaborative filtering recommendation technology. Understand the principle and method of recommended attack. Analysis the principle of several attack models such as random attack, and on the basis of the collaborative filtering recommendation algorithm to implement these attack models.The experimental results show the impact of recommended attack on the system.2. Studying several attack detection algorithms including the RDMA, WDMA and FMTD on the basis of understanding the recommended attack. The principle of the detection algorithms are being analyzed. Implementation the detection algorithm to detect the recommended attack.Analyzed and compared the results of the detection algorithms.3.Studying and understanding the UnRAP detection algorithm.In order to improve the precision of detection.On the basis of the UnRAP algorithm to propose two improved detection algorithms by modifying the algorithms of RDMA (Rating Deviation from Mean Agreement) and WDMA (Weighted Deviation form Mean Agreement). 4.Combined with UnRap algorithm, as well as the improved RDMA and WDMA algorithm we propose a hybrid decision approach to effectively and efficiently detect the profile injection attacks in collaborative recommender systems.This algorithm combining multiple detection algorithm for joint decision-making to improve the accuracy and effect of the attack detection. Finally, the experiments be designed to verify the effectiveness of the algorithm by the recommended attack detection.