Spam network discovery

Alfred Hero
University of Michigan

I will present a spectral clustering approach to discovery of dependency networks of email spammer activity. For specific similarity measures the approach can be related to the problem of estimation of a latent variable in a multivariate statistical model. We illustrate our methods on honeypot data collected from the Unspam project.

Presentation (PDF File)

Back to Workshop I: Multiscale Representation, Analysis and Modeling of Internet Data and Measurements