Distance-Based Network Partitioning

TL;DR

This paper introduces a distance-based method for detecting communities in networks by using reference nodes and distance vectors to identify well-separated clusters representing communities at multiple scales.

Contribution

It presents a novel community detection approach that uses reference nodes and distance vectors to define an interpretable coordinate system for network partitioning.

Findings

Effective in identifying communities in benchmark networks

Applicable to spatial and city street networks

Provides intuitive interpretation of node-community relationships

Abstract

A new method for identifying communities in networks is proposed. Reference nodes, either selected using a priory information about the network or according to relevant node measurements, are obtained so as to indicate putative communities. Distance vectors between each network node and the reference nodes are then used for defining a coordinate system representing the community structure of the network at many different scales. For modular networks, the distribution of nodes in this space often results in a well-separated clustered structure, with each cluster corresponding to a community. One interesting feature of the reported methodology for community finding is that the coordinate system defined by the seeds allows an intuitive and direct interpretation of the situation of each node with respect to the considered communities. The potential of the method is illustrated with respect to a community detection benchmark, a spatial network model and to city streets networks.