On the Hyperbolicity of Large-Scale Networks

TL;DR

This paper investigates the hyperbolicity of large-scale networks, demonstrating that social and communication networks are hyperbolic while road networks are not, and introduces scalable methods for detecting hyperbolicity in big graphs.

Contribution

The study provides extensive empirical evidence that large-scale networks are generally hyperbolic and introduces efficient renormalization techniques for hyperbolicity detection in large graphs.

Findings

Social and communication networks exhibit hyperbolicity.

Road networks are not hyperbolic.

Renormalization preserves and amplifies hyperbolicity.

Abstract

Through detailed analysis of scores of publicly available data sets corresponding to a wide range of large-scale networks, from communication and road networks to various forms of social networks, we explore a little-studied geometric characteristic of real-life networks, namely their hyperbolicity. In smooth geometry, hyperbolicity captures the notion of negative curvature; within the more abstract context of metric spaces, it can be generalized as d-hyperbolicity. This generalized definition can be applied to graphs, which we explore in this report. We provide strong evidence that communication and social networks exhibit this fundamental property, and through extensive computations we quantify the degree of hyperbolicity of each network in comparison to its diameter. By contrast, and as evidence of the validity of the methodology, applying the same methods to the road networks shows that they are not hyperbolic, which is as expected. Finally, we present practical computational means for detection of hyperbolicity and show how the test itself may be scaled to much larger graphs than those we examined via renormalization group methodology. Using well-understood mechanisms, we provide evidence through synthetically generated graphs that hyperbolicity is preserved and indeed amplified by renormalization. This allows us to detect hyperbolicity in large networks efficiently, through much smaller renormalized versions. These observations indicate that d-hyperbolicity is a common feature of large-scale networks. We propose that d-hyperbolicity in conjunction with other local characteristics of networks, such as the degree distribution and clustering coefficients, provide a more complete unifying picture of networks, and helps classify in a parsimonious way what is otherwise a bewildering and complex array of features and characteristics specific to each natural and man-made network.