Research On Attack Profile Detection Algorithm For Collaborative Filtering Recommender Systems

Collaborative filtering recommendation system provides an effective way to solvethe problem of“information overload”, and has a wide range of applications ine-commerce. However, due to the openness and user-anonymity, there are serious hiddendangers in the recommend system. Some malicious users can provide a number of userprofiles to the system for some commerce purposes, trying to make the system productadvantage recommendation for them. This can make a serious impact on therecommendation quality of the system. So, how to solve the security problem ofrecommended system is urgently needed. On the basis of the research and analysis aboutthe present situation at home and abroad，this paper has further conducted deep researchof the security problem of recommended system.Firstly, from the characteristics of user profile and the difference among the rating oftarget items, according to the problem that the existing attack profile detection algorithmscan not successfully detect multiple items under attack at the same time, an unsupervisedalgorithm for detection of attack profile based on multi-target items retrieval is proposed.We firstly look for suspicious target items in the user profile subsets, and dynamicallygenerate the target item tree, then identify the target items under attack and theircorresponding attack profiles according to the coalitional rating deviation of the items.Secondly, from the degree of user profiles from away their nearest, according to theproblem that the existing user profile attack detection algorithms have a low detection pr-ecision about obfuscated attack, an unsupervised algorithm for detecting attack profiles b-ased on local density is proposed. We firstly calculate the local outlier factor for each userprofile and get the local deviation degree of the users. Then combined with the differenceof the target item rated by attackers and genuine users, we can find the target item and at-tack purpose so as to identify the corresponding attack profiles. The experimental resultsshow that the proposed algorithm has higher accuracy both in detecting standard attack a-nd obfuscated attack.Finally, we give the experimental evaluations and analysis of the algorithms proposed in this paper, compare the performance between the proposed algorithms andother existing algorithms, and make the conclusions and prospects for the furthersearch.