Design and implementation of a privacy-enhanced microblogging architecture

Abstract

Microblogging services like Twitter allow for large-scale information sharing and retrieval. They are used by millions of people and have become a tremendously popular platform to share both inconsequential as well as sensitive information such as political views, habits or health conditions. These services are designed and geared towards allowing easy and straightforward sharing of content, but they often overlook the privacy of this information. While some services allow a somewhat granular access control to information shared, there remains the problem that the services providers themselves have unlimited access to the information of their user base. This information consists both of the actual content shared as well as the interests of users in topics and publishers. Users have to trust the service provider that they do not misuse this wealth of information. While they are usually bound by reputation to safeguard it, there are often clauses in user agreements that allow providers to mine user content and deliver targeted advertisements or to resell information to their partners. In previous work some of these issues have been addressed by encrypting the shared content as well as the interests in specific topics. This drastically limits the information gained by the provider, but it still retains the information of whom/what users are interested in; i.e. user relationships. In this master thesis we try to circumvent these problems by designing and implementing a Twitter-like microblogging service that gains no information at all about user relationships. Towards this goal we use a technique called Private Information Retrieval (PIR) which enables a user to retrieve data from a database without the database host gaining any information about what he retrieved. We use a practical approach to PIR and design and implement a fully functional microblogging system which makes use of the provider’s cloud and is user-friendly by doing most of the client-side work with an appropriate browser extension. It is general enough such that it can be extended by other proposed systems for further privacy through encryption of the shared content. We evaluate our system in a real cloud, discuss the results and draw conclusions about the feasibility of using PIR in a scalable, dynamic system.