Table 3 Results of technique validation experiment.

From: An Open Graph Dataset Organized by Scales

Distance Metrics

Kolmogorov-Smirnov Distance (KSD)

Skew Divergence Distance (SDD)

L2-normalization Distance (L2ND)

 

Fine-tune*

RE

RNE

TIES

Fine-tune*

RE

RNE

TIES

Fine-tune*

RE

RNE

TIES

Degree Distribution

0.138  ± 0.014

0.480  ± 0.036

0.142  ± 0.026

0.153  ± 0.014

0.008  ± 0.002

0.023  ± 0.003

0.008  ± 0.002

0.005  ± 0.001

0.052  ± 0.01

0.153  ± 0.014

0.048  ± 0.01

0.092  ± 0.006

Clustering Coefficient

0.072  ± 0.019

0.332  ± 0.039

0.077  ± 0.019

0.077  ± 0.011

0.005  ± 0.002

0.019  ± 0.003

0.006  ± 0.002

0.003  ± 0.001

0.036  ± 0.009

0.127  ± 0.015

0.034  ± 0.009

0.082  ± 0.007

PageRank Distribution

0.137  ± 0.019

0.415  ± 0.024

0.154  ± 0.024

0.420  ± 0.026

0.011  ± 0.002

0.011  ± 0.002

0.018  ± 0.002

0.019  ± 0.003

0.065  ± 0.012

0.161  ± 0.009

0.064  ± 0.012

0.163 ± 0.011

  1. Mean KSD, SDD, and L2ND values with standard error (95% confidence intervals) for degree, clustering coefficient, and PageRank distributions measured across all 77 fine-tuned scale levels. For our fine-tuning approach (indicated by “*”), structural metrics were computed between the original and fine-tuned graphs for each level. For the reference methods, the original simple graph was sampled five times per scale level (with the edge count adjusted to match that of the fine-tuned version), and the metrics were computed in the same way. Values close to zero indicate high similarity between the distributions.
Back to article page