A privacy-based mechanism for users' information scoring and anonymisation across multiple online social networks

Social network sites are becoming more and more popular among individuals in recent years and have eased social interactions to help individuals connect with others with a common interest, and to exchange information. As individuals share their personal information such as age, job details, views, opinions and thoughts on such sites, they may face different privacy issues such as identity theft, bullying, harassment and even job termination. As the participation of users in social networking sites increases, the likelihood of sharing information with unknown users escalates, and the possibility of privacy risks for the user is elevated. There are two main ways suggested in the literature to minimise the privacy risks of users on social media sites. The first is to measure the privacy risk. The second is to hide sensitive information from others. To measure the privacy risk, there are several studies on scoring privacy for online social media users for structured data (data contained in fields such as name, age and qualification) in a single source, neglecting the fact that social media users, in general, have multiple social network profiles revealing dissimilar sensitive information which can aggravate the risk. Moreover, there are limited works on privacy calculation of the risk caused by unstructured data (any textual data). For preserving the privacy of data, several anonymisation techniques have been proposed. However, in the context of preserving the privacy of individuals during the friending phase (the act of adding someone as a friend) in social media, there are only a few available approaches. Most of them disregard the privacy from the user's perspective and are more focused on users' security rather than privacy. To address these problems, this thesis proposes approaches that can support online social network users to quantify their privacy disclosure based on their structured and unstructured information shared across multiple social media sites. Evaluation of the study illustrates that the proposed models can deliver a better approximation of privacy for users with multiple profiles on online social networks. This thesis also investigates a privacy-preserving friending method for information sharing across multiple social media sites. As friending exposes the sensitive data of a user to others, this model helps individuals to decide how to share their information safely through social networking sites with a reduced risk of being exploited or re-identified. Evaluation of the model shows that information sensitivity calculation, as well as anonymisation, offers a more effective way of friending. The key research findings and contributions of this thesis are: ‚Äö Despite several governmental and social networks policy changes, privacy risk is still a significant problem with social media sites. ‚Äö It is important to consider the visibility and sensitivity of the information on multiple online social network sites to improve the accuracy of privacy risk evaluation. ‚Äö For accurate and credible scoring the privacy risk of unstructured data such as tweets, blogs and comments, and structured information should also be considered along with sentiment associated with unstructured data. ‚Äö By considering the sensitivity of the information in anonymisation of shared data, privacy-preserved friending can be achieved with reduced privacy risks for social media users.