Desarrollo de un modelo de recomendación grupal que considere la privacidad de los usuarios usando datos sintéticos y agregación de preferencias.

Abstract

In this work, we propose a group recommendation model is proposed it is shown that using synthetic data, data protection is maintained. For this, two approaches were used to generate synthetic data and thus be able to hide the private information of the group’s users. The first one is ≪differential privacy≫ which uses a Bayesian network model that combines lowdimensional distributions to approximate the full-dimensional distribution of a data set. The second approach used was the ≪CART≫ method which uses sequential models and applies a transformation to the original data in such a way that only some values are changed. The results with the synthetic data of both approaches are quite promising and it is verified that they manipulate less specific information about the preferences of the users in the group. Our work also shows that it is possible to guarantee data privacy without a significant loss of accuracy in group recommendation.