Prevention of Shilling Attack in Recommender Systems using Discrete Wavelet Transform and Support Vector Machine

Recommender systems that work by collaborative filtering technique are prone to shilling attack. "Shilling" or "profile injection" attacks are known to occur when an individual or an organization is trying to promote an item. They aim to manipulate the working of recommender system by injecting a large number of fake profiles into it so that the output of the system becomes biased. Most of the previous works rely on batch processing technique which is tedious. Hence, we propose an online detection method. We propose a hybrid approach by combining Discrete Wavelet Transform (DWT) and Support Vector Machine (SVM) to classify the fake profiles. The proposed system aims to construct rating series of individual users from the popularity and novelty of items they have rated to classify them as fake or genuine users. Discrete Wavelet Transform has been applied on the rating series to get the feature set, which is then used by Support Vector Machine for classification. All experiments are done on MovieLens 100K dataset. The results show that DWT-SVM performs better compared with HHT-SVM (Hilbert Huang Transform-SVM), PCA-VarSelect (Principal Component Analysis-Variable Selection), and Batch SVM methods.