Detecting shilling profiles in collaborative recommender systems via multidimensional profile temporal features
IET Information Security
To defend recommender systems, various methods have been proposed to detect shilling profiles, which can be categorised as user-and item-based detection methods. Most of the user-based methods identify shilling profiles via statistical signatures of rating values and suffer from low precision when detecting different types of attacks. Most of the item-based methods use temporal information to detect the anomaly items, but they assume that the fake ratings were injected in short periods. So they
... are invalid for the long duration and decentralised injection attacks. To address these limitations, the authors extract the multidimensional profile temporal features and present a shilling detection method. First, from the user profile view, user rating behaviours are characterised by corrected conditional entropy and the dissimilarity with the rest-rating model. Second, from the item profile view, the user features are extracted according to item temporal popularity. Third, the features based on weighted deviation from dynamic mean are extracted according to the fact that the items mean changes with time. Finally, support vector machine is exploited to detect shilling profiles based on the proposed features. Experimental results on the Netflix dataset indicate that the performance of the proposed method is better than that of the benchmark methods. Nomenclature U set of the entire users U g set of genuine users in the training set I set of the entire items I u set of items rated by user u D rating database, including UserId, ItemId, rating, and timestamp r u, i, t rating of user u to item i at t moment r u, i, t ≠ ⊥ item i is rated by user u at t moment r u, i, t = ⊥ item i is not rated by user u at t moment ⋅ the cardinality of the set Γ(r u, i, t ) discriminate function, if r u, i, t ≠ ⊥, the value is 1; otherwise the value is 0 IET Inf. Secur. © The Institution of Engineering and Technology 2018 1 iii. Based on the multidimensional profile temporal features, we train an SVM classifier to detect shilling profiles and conduct experiments on the Netflix contest dataset to demonstrate its effectiveness. The rest of this paper is organised as follows. Section 2 briefly discusses the related work. Section 3 introduces the definition of the notations and some shilling attack models. In Section 4, we describe the detection method, the feature extraction method, and the algorithm. Next, in Section 5, we conduct experiments and analyse the performance of the proposed method. Finally, the conclusions are drawn in Section 6. Fig. 8 Recall of six methods with four attack types at various filler sizes across various attack sizes. Note that the recall of PCA-VarSelect and DegreeSAD under AoP attack is 0 except for the recall of PCA-VarSelect at 12% attack size (a) 3% filler size, (b) 5% filler size IET Inf. Secur. © The Institution of Engineering and Technology 2018 9 Fig. 11 Comparison of PS values for the matrix factorisation algorithms that include using SVM-MPTF to perform shilling detection and no shilling detection under four attacks with various filler sizes and attack sizes 12 IET Inf. Secur.