On the Scale-Free Property of Citation Networks: An Empirical Study

Citation networks have been thought to exhibit scale-free property for many years; however, this assertion has been doubted recently. In this paper, we conduct extensive experiments to resolve this controversial issue. We firstly demonstrate the scale-free property in scale-free networks sampled from the popular Barabasi-Albert (BA) model. To this end, we employ a merged rank distribution, which is divided into outliers, power-law segment, and non-power-law data, to characterize network degrees, and propose a random sample consensus (RANSAC)-based method to identify power-law segments from merged rank distributions, and use the Kolmogorov-Smirnov (KS) test to examine the scale-free property in power-law segments. Subsequently, we apply the same methods to examine the scale-free property in real-world citation networks. Experimental results confirm the scale-free property in citation networks and attribute previous skepticism to the presence of outliers.