Privacy-preserving multi-criteria collaborative filtering

Privacy-preserving collaborative filtering systems focus on eliminating the privacy threats inherent in single preference values, and the privacy risks in the multi-criteria preference domain are disregarded. The structure of multi-criteria preference data exposes individuals to more severe privacy threats although it provides the opportunity to understand why an item is preferred by the user. Therefore, these systems require intelligent protection mechanisms that are flexible and adapting to the structure of each sub-criterion. In this dissertation, existing privacy violation conditions from the perspective of multi-criteria recommender systems are evaluated and threats exposed by such services are discussed comprehensively. In order to alleviate such threats, randomized perturbation-based privacy-preserving approaches for multi-criteria collaborative filtering systems and the privacy protection methods efficiently used in traditional single-criterion systems are adapted onto multi-criteria ratings. To increase the prediction accuracy, a novel privacy-preserving protocol by adapting an entropy-based randomness determination procedure is introduced that can recover accuracy losses resulting from perturbation of original multi-criteria preferences. In addition, a novel data perturbation approach was introduced to mitigate the adverse effects of unusual user ratings on prediction accuracy. The proposed schemes are experimentally evaluated on three subsets of Yahoo!Movies multi-criteria preference dataset to demonstrate the effects of the proposed privacy-preserving schemes on both user privacy levels and prediction accuracy for differing sparsity rates. According to the obtained experimental outcomes, the proposed schemes can produce significantly more accurate predictions while maintaining an identical level of privacy provided by the traditional privacy protection scenario.