Separating Terrorist-Like Topological Signatures Embedded in Benign Networks

Kshiteesh Hegde, Malik Magdon-Ismail

Abstract: We study the problem of identifying topologically adversarial nodes in real networks using a graph classification methodology. To test our approach, we implant nodes from a terrorist network in a wide variety of benign real networks to create a topologically heterogeneous network. We capture local information using structured image embeddings of adjacency matrices and identify the terrorist nodes. We recover up to 85% of the terrorist-like nodes implanted in friendly host networks while keeping the false positive rate as low as 1%.