The effect of aging on network structure

TL;DR

This paper investigates how aging effects, modeled by exponential and power-law decay factors, influence the structure and properties of evolving networks, revealing significant transformations such as clustering, hierarchy, and changes in average distance.

Contribution

It introduces a framework to incorporate aging effects into network models, demonstrating their impact on network topology and properties, which was not thoroughly explored before.

Findings

Aging induces clustering and hierarchical structures.

Different decay types lead to distinct network transformations.

Aging affects average node distances and network topology.

Abstract

In network evolution, the effect of aging is universal: in scientific collaboration network, scientists have a finite time span of being active; in movie actors network, once popular stars are retiring from stage; devices on the Internet may become outmoded with techniques developing so rapidly. Here we find in citation networks that this effect can be represented by an exponential decay factor, $e^{-\beta \tau}$, where $\tau $ is the node age, while other evolving networks (the Internet for instance) may have different types of aging, for example, a power-law decay factor, which is also studied and compared. It has been found that as soon as such a factor is introduced to the Barabasi-Albert Scale-Free model, the network will be significantly transformed. The network will be clustered even with infinitely large size, and the clustering coefficient varies greatly with the intensity of the aging effect, i.e. it increases linearly with $\beta $ for small values of $\beta $ and decays exponentially for large values of $\beta $. At the same time, the aging effect may also result in a hierarchical structure and a disassortative degree-degree correlation. Generally the aging effect will increase the average distance between nodes, but the result depends on the type of the decay factor. The network appears like a one-dimensional chain when exponential decay is chosen, but with power-law decay, a transformation process is observed, i.e., from a small-world network to a hypercubic lattice, and to a one-dimensional chain finally. The disparities observed for different choices of the decay factor, in clustering, average node distance and probably other aspects not yet identified, are believed to bear significant meaning on empirical data acquisition.