Implementing Spatial Social Networks with Real World Data

Abstract

The three classical models for synthetically generated social networks, Erdos-Renyi, Barabasi-Albert, and Watts-Strogatz have been shown to not reflect real world social networks. Spatial versions of these networks, which take into account the distance between nodes when determining the probability that two nodes connect, have been proposed to improve their realism, but these spatial versions were not tested with real world data. We altered the code for the spatial versions of the three classical models to accommodate real world data, namely the CBG coordinates for Fairfax County. In this project, we evaluate the realism of the networks generated by this code, by collecting descriptive statistics such as the number of triangles and the average length of edges, as real world networks are likely to have high clustering and nodes are more likely to connect to nodes that are near them. Our experimental results show that the networks that were generated by the spatial versions of the three models had a higher clustering coefficient, a higher number of triangles, and a lower average distance between connected nodes than their nonspatial counterparts. The visual representations of the spatial networks also demonstrated clear clustering. This implies that the networks which take distance between nodes into account are more similar to real world networks.