Classifying Online Social Network Users Through the Social Graph

Cristina Perez-Sola (Universitat Autonoma de Barcelona, Spain) and Jordi Herrera-Joancomarti (Universitat Autonoma de Barcelona & Internet Interdisciplinary Institute, Spain)

In this paper, we address the problem of classifying online social network users using a naively anonymized version of a social graph. We use two main user attributes defined by the graph structure to build an initial classifier, node degree and clustering coefficient, and then exploit user relationships to build a second classifier. We describe how to combine these two classifiers to build an OSN user classifier and then we evaluate the performance of our architecture by trying to solve two different classification problems (a binary and a multiclass problem) using data extracted from Twitter. Results show that the proposed classifier is sound and that both classification problems are feasible to solve by an attacker who is able to obtain a naively anonymized version of a social graph.






FPS 2012 Program