Personalization has recently become an influential part of social media, as it significantly affects the ways people interact with each other through the systems. As more and more social media providers automatically personalize their services based on user data, of which process is often black-boxed, it hinders user's autonomy in their use of systems. In this context, this dissertation investigates the ways to design for user autonomy in a system-driven personalization of social media. Through three-phase studies, this dissertation aims to provide a solid set of design principles that researchers of future "smart" social systems can leverage in their practices.