Evolutionary modeling of a blog network

A common approach to produce theory to explain the genesis and dynamics of complex networks is to create multi-agent simulations that output networks with similar characteristics to the ones derived from real data. For example, a well know explanation for the power law degree distributions found in blog (and other) networks is the agent-level endogenous mechanism of preferential attachment. However, once simplifying assumptions are dropped, finding lower level behaviors that explain global network features can become difficult. One case, explored in this paper, is that of modeling a blog network generated by human agents with heterogeneous behaviors and a priori diversity. We propose an approach based on an hybrid strategy, combining a generic behavioral template created by a human designer with a set of programs evolved using genetic programming. We present experimental results that illustrate how this approach can be successfully used to discover a set of non-trivial agent-level behaviors that generate a network that fits observed data. We then use the model to make successful testable predictions about the real data. We analyze the diversity of behaviors found in the evolved model by clustering the agents according to the execution paths their programs take during the simulation. We show that these clusters map to different behaviors, giving credence to the need for exogenous, in addition to the more conventional endogenous, explanations for the dynamics of blog networks.