Osho, Abiola Afolake2022-04-152022-04-15https://hdl.handle.net/2097/42153Due to their wide reach, oversimplified conversations, and ability to provide quick blasts of information, online social networks (OSNs) have become an avenue where users connect to share information, news, and events around the world. Third-party recommender systems, spammers, and manipulators can learn a user’s online behavior and interests through their connections and interactions because users would often leave breadcrumbs on their interests and personality through activities - like, comments, share, repost, etc. With access to user data and sophisticated learning models, manipulation through inferential attack are now easier to achieve, causing users to struggle with privacy loss as a consequence of their participation in OSNs. Given that some users have a higher propensity for disclosure than others, a one-size-fits-all technique for limiting manipulation and privacy loss proves insufficient. In the search to find a balance between privacy preservation and social influence, we propose a solution that uses the information spread behavior as a basis for estimating the possible exposure of users to abuse and misinformation in the network. We focus on the information spread behavior and explore how it can be used for manipulation purposes. We explore a microscopic follower-followee relationship to show how direct interactions can contribute to targeted manipulation based on inferential attack. The proposed model utilizes the user’s probability of engaging with a post as a way to measure their sensitivity to privacy loss. With this knowledge, the user can then implement a privacy preservation mechanism to minimize their privacy loss by adding noise to their profile to muddle up an attacker’s opinion of them. The result from experiments on real-world Twitter data showed that even though there will be costs to participating in OSNs, these costs can be minimized relative to the disclosure threshold set by the user as their maximum privacy loss. Additionally, we report attributes that can be tweaked to minimize the user’s exposure.en-USBehavioral analysisSocial network analysisInformation diffusionMisinformationManipulationAbuseDissertation