AUTOMATIC DISCOVERY OF AGENT BASED MODELS: AN APPLICATION TO SOCIAL ANTHROPOLOGY

We present a methodology that applies a machine learning technique — genetic programming — to the problem of finding plausible generative models for complex networks. We specifically apply this method to the analysis of alliance networks, a type of kinship network used by social anthropologists where nodes are groups and directed edges represent a group giving a wife to another group. Network generators are represented as computer programs. Evolutionary search is used to find programs that generate networks that best approximate real networks. The quality evaluation of a model is based on a set of network metrics with anthropological meaning. We evolve generators for seventeen real alliance networks and find that our approach is capable of generating high quality results both in terms of network similarity and human readability of the programs. We present and discuss a subset of the experimental results that highlights several interesting aspects of our findings. We believe in the applicability of the methodology to complex networks in general and propose that these are the first steps towards an artificial network scientist.