Generate Descriptive Social Networks for Large Populations from Available Observations: A Novel Methodology and a Generator (2003.02213v1)

Abstract: When modeling a social dynamics with an agent-oriented approach, researchers have to describe the structure of interactions within the population. Given the intractability of extensive network collecting, they rely on random network generators that are supposed to explore the space of plausible networks. We first identify the needs of modelers, including placing heterogeneous agents on the network given their attributes and differentiating the various types of social links that lead to different interactions. We point out the existence of data in the form of scattered statistics and qualitative observations, that should be used to parameter the generator. We propose a new approach peculiar to agent-based modeling, in which we will generate social links from individuals' observed attributes, and return them as a multiplex network. Interdependencies between socioeconomic attributes, and generative rules, are encoded as Bayesian networks. A methodology guides modelers through the formalization of these parameters. This approach is illustrated by describing the structure of interactions that supports diffusion of contraceptive solutions in rural Kenya.