Effective Privacy Preservation Models for Rating Datasets

Abstract

Rating datasets are data collections of user prole tuples that generally include their person-specific data and the preferred level of their interesting artifacts. Generally, rating datasets are generated for use by recommendation methods that are available in real-life systems such as mobile applications, location-based services, e-commerce systems, and websites. Aside from recommendation methods, we can see that rating datasets can further be utilized by data analysts with appropriate business reasons such as improving system policies and constructing business reports. Such a data utilization of rating datasets can lead to privacy violation issues. To address these issues in rating datasets, (lp1 , . . . , lpn )-Privacy is proposed. That is, before rating datasets are utilized by the specified recommendation method and data analysts, the user's unique preferences in rating datasets are generalized by less specific data to be indistinguishable. Moreover, the generalized rating dataset can only be utilized through aggregate query frameworks. Although this privacy preservation model can address privacy violation issues in rating datasets, it still has a serious data utility issue that must be improved. For this reason, a new effective privacy preservation model for rating datasets is proposed in this work. Moreover, we propose practical experiments that can measure the effectiveness of the proposed privacy preservation model. The experimental results show that the proposed privacy preservation model is highly effective.