Collaborative filtering recommender systems (CFRSs) are the key components ofsuccessful e-commerce systems. Actually, CFRSs are highly vulnerable to attackssince its openness. However, since attack size is far smaller than that ofgenuine users, conventional supervised learning based detection methods couldbe too "dull" to handle such imbalanced classification. In this paper, weimprove detection performance from following two aspects. First, we extractwell-designed features from user profiles based on the statistical propertiesof the diverse attack models, making hard classification task becomes easier toperform. Then, refer to the general idea of re-scale Boosting (RBoosting) andAdaBoost, we apply a variant of AdaBoost, called the re-scale AdaBoost(RAdaBoost) as our detection method based on extracted features. RAdaBoost iscomparable to the optimal Boosting-type algorithm and can effectively improvethe performance in some hard scenarios. Finally, a series of experiments on theMovieLens-100K data set are conducted to demonstrate the outperformance ofRAdaBoost comparing with some classical techniques such as SVM, kNN andAdaBoost.
展开▼
