Doi: 10.1145/1400181.1400198
SCIENTISTS FROM A variety of disciplines, including physics, sociology, biology, and computing, all explore the topological properties of complex systems that can be characterized as large-scale networks, including scientific collaborations, the Web, the Internet, electric power grids, and biological and social networks. Despite the differences in their components, functions, and size, they are surprisingly similar in topology, leading to the conjecture that many complex systems are governed by the ubiquitous “self-organizing” principle, or that the internal complexity of a system increases without being guided or managed by external sources.
Still missing from this line of research, however, is an analysis of the topology of “dark” networks hidden from view yet that could have devastating effects on our social order and economy. Terrorist organizations, drug-trafficking rings, arms-smuggling operations, gang enterprises, and many other covert networks
are dark networks. Their structures are largely unknown to outsiders due to the difficulty of accessing and collecting reliable data. Do they share the same topological properties as other types of empirical networks? Do they follow the self-organizing principle? How do they achieve efficiency under constant surveillance and threat from the authorities? How robust are they against attack? Here, we explore the topological properties of several covert criminal- and terrorist-related networks, hoping to contribute to the general understanding of the structural properties of complex systems in hostile environments while providing authorities insight regarding disruptive strategies.
Topological analysis focusing on the statistical characteristics of network structure is a relatively new methodology for studying large-scale networks. 1, 11 Large complex networks can be categorized into three types: random, small-world, and scale-free. 1 A number of statistics (see Table 1) have been developed to study their topology; three of which—average path length, average clustering coefficient, and degree distribution—are widely used to categorize networks.
In random networks, two arbitrary nodes are connected with a probability p; as a result each node has roughly the same number of links. Random networks are characterized by small l, small C, and bell-shaped Poisson distributions. 1 A small l means an arbitrary node can reach any other node in a few steps. A small C implies that random networks are not likely to contain clusters and groups. Studies by physicists and computer and social scientists have found that most complex systems are not random but present small-world and scale-free properties (see Albert1 for a comprehensive review of these studies).
A small-world network has a significantly larger C than its random-network counterpart while maintaining a relatively small l. 11 Scale-free networks, on the other hand, are char-
ILLUS TRATION B Y LEANDER HERZOG
References:
Archives