Supervised approach for detecting average over popular items attack in collaborative recommender systems

摘要: Recent research has shown the significant vulnerabilities of collaborative recommender systems in face profile injection attacks, which malicious users insert fake profiles into rating database order to bias system's output. To reduce this risk, a number approaches have been proposed detect such attacks. Although existing detection can standard type these attacks effectively, they perform badly when detecting recently obfuscated for example, average over popular items (AoP) attack. With problem mind, study author propose supervised approach First, he uses theory term frequency inverse document (TFIDF) extract features AoP Second, training set train support vector machine (SVM) generate SVM-based classifier. Finally, generated classifier The experimental results on MovieLens dataset show that attack with high recall and precision.