On the Structural Properties of Social Networks and their Measurement-calibrated Synthetic Counterparts

TL;DR

This study analyzes the structural properties of 120 real social networks and their synthetic counterparts, revealing domain-dependent correlation patterns and evaluating the models' ability to replicate key topological features.

Contribution

It provides a comprehensive comparison of real and synthetic social networks, identifying domain-specific differences and limitations of common network models.

Findings

Correlation patterns differ across social domains

A non-redundant set of metrics to describe social networks

Model fit varies depending on the social domain

Abstract

Data-driven analysis of large social networks has attracted a great deal of research interest. In this paper, we investigate 120 real social networks and their measurement-calibrated synthetic counterparts generated by four well-known network models. We investigate the structural properties of the networks revealing the correlation profiles of graph metrics across various social domains (friendship networks, communication networks, and collaboration networks). We find that the correlation patterns differ across domains. We identify a non-redundant set of metrics to describe social networks. We study which topological characteristics of real networks the models can or cannot capture. We find that the goodness-of-fit of the network models depends on the domains. Furthermore, while 2K and stochastic block models lack the capability of generating graphs with large diameter and high clustering coefficient at the same time, they can still be used to mimic social networks relatively efficiently.