|
Abstract:
|
In many applications, particularly, on analysis of various dark (i.e., criminal, terrorist and illicit) networks, there often exists some prior knowledge on node labels, and the primary interest is to cluster the remaining network data, given this prior set of labelled training data. In this talk we propose a new nonparametric supervised algorithm for detecting multiple communities in complex networks using the Depth vs. Depth (DD(G)) classifier. The proposed new DD(G)-method is inherently geometric and allows to simultaneously account for network communities and outliers. We illustrate utility of the new approach in application to analysis of structure in terrorist and extremist organizations and their interactions within the United States.
|
