Enhancing privacy approach of recommender systems

Abstract

In the last few decades, recommendation systems has received an iconic representation in the field of information technology. With the noticed rapid advancement of data mining, the issue of privacy has become an inevitable necessity. Hence, the mainstream challenge that accompanies data mining is developing a cutting-edge strategy to protect private information. In this work, we present two frameworks for enhancing and preserving the privacy of data in recommendation systems along with their experimental results and discussions. In the first framework, we proposed a hybrid strategy data perturbation and query restriction (DPQR) with an improved version of MASK (Mining association rules with secrecy Konstraints) scheme to decrease the complexity of traditional MASK scheme. This hybridization resulted in 49.7% as recommendation precision and privacy degree of 97.4% while the traditional MASK scheme gives only 80% privacy degree. We enhanced our results by adopting non-linear programing and solved the privacy problem as a system of equations by setting the privacy equation to be our objective function, the privacy degree was raised to 99.6% and the recommendation precision reached 59%.