Computational Methods for Testing Adequacy and Quality of Massive Synthetic Proximity Social Networks

Abstract

Social contact networks play an increasingly important role in computational epidemiology. We focus on massive social contact networks that cover an urban region comprising of millions of individuals and billions of time varying interactions. It is impossible to obtain such networks by simply measuring social interactions. As a result, such networks are often synthesized using a diverse set of real-world data. The synthesis method can be viewed as a complex stochastic process that outputs one realization of such a network. The resulting networks are extremely large, dynamic and unstructured. Any meaningful description of such networks is usually done in terms of structural properties. Building on our earlier work, we synthesize a detailed social contact network for the National Capital Territory (NCT) of India. We first synthesize a social visitation network, representing people visiting locations during different time intervals. We then project it to synthesize a people-people contact network. We are not aware of other works on synthesizing the NCT network. Two important questions arise when synthesizing such massive dynamic social contact networks: (i) how does one compare the networks that span the same region and (ii) when is the synthesized network adequate. To address them, we compute a number of network measurements: some of which are classical, while others capture the semantics of social contact networks. These metrics are used to study the similarities and differences between two networks representing the same urban region. For question (ii), we study our ability to understand the dynamics and control of epidemics. Dynamical measures that capture the joint interaction between the local dynamical process and the network structure are presented and used to analyze the NCT networks.