Efficient Anonymization of Vulnerable Individuals in Social Networks
Social networks, patient networks, and email networks are all examples of graphs that can be studied to learn about information diffusion, community structure, and different system processes; however, they are also all examples of graphs containing potentially sensitive information. While several anonymization techniques have been proposed for social network data publishing, they all apply the anonymization procedure on the entire graph. This thesis proposes a local anonymization algorithm that focuses on obscuring structurally important nodes that are not well anonymized. Doing so reduces the cost of the overall anonymization procedure. Based on our experiments, we observe that we reduce the cost of anonymization by an order of magnitude while maintaining, and even improving, the accuracy of different graph centrality measures, e.g. degree and betweenness, when compared to another well-known data publishing approach. This thesis then explores the underlying anonymity inherent in the topological structure of online social networks to better understand which parts are not well anonymized. We find that while some components are well anonymized, subgraphs of weak nodes do exist. The final part of this thesis then extends a known measure for assessing the level of anonymization in a network to consider nodes that are more important than others. We find that this measure better captures the potential impact of nodes in the network being identified by an attacker.
MetadataShow full item record
Showing items related by title, author, creator and subject.
Research With the Doubly Vulnerable Population of Individuals Who Abuse Alcohol: An Ethical Dilemma Gwyn, Priscilla Gage; Colin, Jessie M. (2010-02)Individuals who have maladaptive patterns of drinking alcohol fall into the category of vulnerable research participants for many reasons, not the least of which includes the stigma often placed on individuals who abuse ...