CSE Speaker Series – Lucas Stuyvesant Thesis Defense

On Wednesday, August 15th at 10:00 am in Cramer 221, Lucas Stuyvesant will defend his thesis “Real-Time Identity Classification on Behavior Modeled Dynamic Social Networks”.


Criminals and terrorists take measures to conceal their identity, including in the digital domain. Existing strategies to identify them have several limitations against a clever adversary, such as the persistent dependence on a targeted user of known identity. If this user can re-anonymize themselves, they are able to stop many of these strategies until the target can be re-identified. However, these strategies do not address the problem of re-identifying users, possibly requiring more expensive or dangerous techniques to be used.

In this thesis, the use of machine learning on behavior models of actors in dynamic social networks were used to demonstrate the capability to re-identify anonymized users. These solutions rely on the lack of anonymity in real-time social networks and the inapplicability of anonymization techniques from the published social network domain. These solutions were designed to use metrics of actors in a social network and evaluated against two actor behavior models, resulting in maximum average true positive rates of 89.7\% and 78.9\% on a sample of actors. These results show that it is possible to use social network metrics to deanonymize a user in real-time using a trained machine learning model.

Posted on .