DLPDS: Learning Users’ Information Sharing Behaviors for Privacy Default Setting in Recommender System

The proliferation of Internet of things has allowed users to provide preference feedback and maintain profiles in multiple websites, which could indicate tastes covering several kinds of domains but focus on small number of topics. Leveraging all the user information available in several sites or domains may be beneficial for knowing the users better and generating higher-quality recommendations. However, aggregating all users’ information globally could trigger users’ awareness on privacy concerns, which further cause users refuse to share the information so that the recommendation quality is reduced. We provide evidence that a recommender system could mitigate the privacy-integration problem, when it is applied with our novel model, called disclosure-learning privacy default setting (DLPDS), which transfer the pattern of users’ past information sharing behaviors into privacy default settings. The result of the experiment support that our DLPDS model could gain users’ trust and aggregate more users information, and that adapting the privacy default settings to the user information sharing pattern may results in positive feedback that promoting better prediction accuracy of the recommender system.