Doctor of Philosophy (Ph.D.)
With billions of users, Online Social Networks(OSNs) are amongst the largest scale communication applications on the Internet. OSNs enable users to easily access news from local and worldwide, as well as share information publicly and interact with friends. On the negative side, OSNs are also abused by spammers to distribute ads or malicious information, such as scams, fraud, and even manipulate public political opinions. Having achieved significant commercial success with large amount of user information, OSNs do treat the security and privacy of their users seriously and provide several mechanisms to reinforce their account security and information privacy. However, the efficacy of those measures is either not thoroughly validated or in need to be improved. In sight of cyber criminals and potential privacy threats on OSNs, we focus on the evaluations and improvements of OSN user privacy configurations, account security protection mechanisms, and trending topic security in this dissertation. We first examine the effectiveness of OSN privacy settings on protecting user privacy. Given each privacy configuration, we propose a corresponding scheme to reveal the target user's basic profile and connection information starting from some leaked connections on the user's homepage. Based on the dataset we collected on Facebook, we calculate the privacy exposure in each privacy setting type and measure the accuracy of our privacy inference schemes with different amount of public information. The evaluation results show that (1) a user's private basic profile can be inferred with high accuracy and (2) connections can be revealed in a significant portion based on even a small number of directly leaked connections. Secondly, we propose a behavioral-profile-based method to detect OSN user account compromisation in a timely manner. Specifically, we propose eight behavioral features to portray a user's social behavior. A user's statistical distributions of those feature values comprise its behavioral profile. Based on the sample data we collected from Facebook, we observe that each user's activities are highly likely to conform to its behavioral profile while two different user's profile tend to diverge from each other, which can be employed for compromisation detection. The evaluation result shows that the more complete and accurate a user's behavioral profile can be built the more accurately compromisation can be detected. Finally, we investigate the manipulation of OSN trending topics. Based on the dataset we collected from Twitter, we manifest the manipulation of trending and a suspect spamming infrastructure. We then measure how accurately the five factors (popularity, coverage, transmission, potential coverage, and reputation) can predict trending using an SVM classifier. We further study the interaction patterns between authenticated accounts and malicious accounts in trending. at last we demonstrate the threats of compromised accounts and sybil accounts to trending through simulation and discuss countermeasures against trending manipulation.
© The Author
Ruan, Xin, "Rethinking Privacy and Security Mechanisms in Online Social Networks" (2016). Dissertations, Theses, and Masters Projects. William & Mary. Paper 1499449857.