In addition to memory- and model-based recommendation schemes, hybrid methods are widely used due to their advantages. Such schemes should be able to provide accurate predictions efficiently while preserving privacy. Also, they need to be robust against possible profile injection or shilling attacks. Although some privacy-preserving memory- and model-based collaborative filtering algorithms have been investigated with respect to robustness, privacy-preserving hybrid recommendation schemes have not been analyzed in terms of robustness.
In this paper, we analyze a privacy-preserving hybrid collaborative filtering scheme with respect to robustness. Four push and two nuke attack models are applied to the algorithm in order to show how robust it is against such shilling attacks. Different sets of experiments are conducted using real data to show how varying controlling parameters affect the robustness. The hybrid scheme is compared with memory- and model-based scheme in terms of robustness. Our analysis show that although the scheme can be marginally considered as robust algorithm, it is less robust than memory- or model-based prediction algorithms with privacy.
Primary Language | English |
---|---|
Journal Section | Articles |
Authors | |
Publication Date | March 31, 2015 |
Submission Date | January 30, 2016 |
Published in Issue | Year 2015 Volume: 4 Issue: 1 |