Especially after the recent privacy scandals in social networks, privacy is getting more and more important to users. Although most users claim to value privacy, their online behavior speaks differently: Most of the privacy settings in their online environment, like social networks, or location sharing services, remain untouched and are not adapted to their privacy needs. In this thesis, I will present an approach to tackle this problem, based on two different pillars. The first part focuses on assisting users in choosing their privacy settings, by using machine learning to derive the optimal set of privacy settings for the user. In contrast to other work, our approach uses context factors as well as individual factors to provide a personalized set of privacy settings. The second part consists of a set of intelligent user interfaces to assist the users throughout the complete privacy journey, from defining friend groups allow targeted information sharing; through user interfaces for selecting information recipients, to find possible errors or unusual settings, and to refine them; up to mechanisms to gather in-situ feedback on privacy incidents, and investigating how to use these to improve a user's privacy in the future. Our studies have shown that including tailoring the privacy settings significantly increases the correctness of the predicted privacy settings; whereas the user interfaces have been shown to significantly decrease the amount of errors, especially unwanted disclosures, that are made when sharing information.