An Empirical Assessment of the Complexity and Realism of Synthetic Social Contact Networks

Abstract

We use multiple measures of graph complexity to evaluate the realism of synthetically-generated networks of human activity, in comparison with several stylized network models as well as a collection of empirical networks from the literature. The synthetic networks are generated by integrating data about human populations from several sources, including the Census, transportation surveys, and geographical data. The resulting networks represent an approximation of daily or weekly human interaction. Our results indicate that the synthetically generated graphs according to our methodology are closer to the real world graphs, as measured across multiple structural measures, than a range of stylized graphs generated using common network models from the literature.

Structural Equivalence in Subgraph Matching

“…For the transportation and twitter networks, the template is extracted from the world graph. The synthetically generated datasets are examples which represent emails, phone calls, financial transactions, among other interactions between individuals and are all generated as part of the DARPA-MAA program [33]- [35]. The subgraph isomorphisms to be detected may be a group of actors involved in adversarial activities including human trafficking and money laundering.…”

“…The multiplex datasets are much larger than the singlechannel graphs in the previous section, with the largest world graphs having hundreds of thousands of nodes and hundreds of millions of edges. The synthetic datasets are divided into three groups based on which organization generated the dataset: PNNL [34], GORDIAN [35], and IvySys Technologies [33]. For our experiments, we examine the same seven modes of equivalence used in the single channel case, but with a time limit of one hour to count as many solutions as possible.…”

“…For these instances, the counting problem is almost entirely solved after applying the initial filter and the solution space is understood by equivalence in the template. For example, observe the template displayed in 2) GORDIAN: The GORDIAN datasets [35] have a much larger templates and worlds than PNNL and they are generated separately in an agent-based fashion to match the daily routines and travel patterns of a certain population of people. Only the FE method fully enumerates the solution space, but the NC and CE methods come close to a full enumeration.…”

Subgraph Matching on Multiplex Networks

“…GORDIAN Version 7 [31] has two instances: Batch-1 and Batch-2. Despite the distinction in sizes and structures of their respective templates, both instances can be solved with node-level statistics, topology and repeated-sets filters.…”