Genetic Algorithm-Based Privacy Preserving Collaborative Filtering

Abstract

Collaborative filtering has become one of the most effective recommender system techniques for generating predictions. However, these systems typically overlook the privacy of individuals as they collect plain preferences of users. Privacy-preserving collaborative filtering systems propose ways of offering predictions without violating the confidentiality of users. Disguising user preferences reduces the accuracy as the data disguising process causes a level of distortion in the preferences. In this study, a genetic algorithm-based approach is proposed to maintain more accurate predictions based on disguised user data. Similarity calculation is improved compared to traditional privacy-preserving collaborative filtering by implementation of genetic algorithm onto a model-based scheme. Experimental outcomes obtained on a real-world data set confirm that the improved genetic algorithm-based scheme significantly outperforms traditional methods in terms of accuracy. © 2019 IEEE.