What is it about?

The study of large ensembles of networks has become a topic of fundamental importance in science and technology. A key component of most network-valued machine learning algorithms is the ability to assign a similarity (or conversely, a distance) score between two networks. This paper provides an extensive review of existing distances between networks.

Featured Image

Why is it important?

This review is significant because it combines experimental results with theoretical analysis of the performance of various network distances. We offer recommendations for the practitioner as well as a Python package that implements all the distances evaluated in the paper.


The comparison of networks often requires a distance to quantify the dissimilarity of the networks. Because the optimal distance will always depend on the specific application, one needs a guide to quantify the performance of existing distances. The evaluation guide should combine a theoretical analysis, with controlled experiments on ensemble of random graphs that capture universal graph structures – in essence the building blocks of existing real world networks.

François Meyer
University of Colorado Boulder

Read the Original

This page is a summary of: Metrics for graph comparison: A practitioner’s guide, PLoS ONE, February 2020, PLOS,
DOI: 10.1371/journal.pone.0228728.
You can read the full text:

Open access logo



The following have contributed to this page