Layered Label Propagation Algorithm

Description

The update rule distinguishes Layered Label Propagation from Label Propagation. Instead of limiting the community choice to the one that is predominant among the nodes in the neighborhood, it takes into account a factor that considers the labeled nodes in the complete network. In fact, both algorithm versions are equivalent whenever this factor is considered 0.

Optimizing equation:

Usage

Install package using NPM.

npm i --save layered-label-propagation

Require it using Node.js.

const llp = require('layered-label-propagation');

Start community finding.

let node2com = llp.jLayeredLabelPropagation(nodes, links, gamma, max);
 
// node2com = {nodeID1: commmunityID1, nodeID2: commmunityID2...}
// nodes = [nodeID1, nodeID2...]
// links = [{source: nodeID1, target: nodeID2, value: weight ...}]
// gamma 
// Whenever the number of iterations exceeds a given "max", algorithms stops further partitioning of the network. 

More

Community Finding with Applications on Phylogenetic Networks (Master Thesis)

Louvain, Infomap, Layered Label Propagation, Label Propagation, Hamming Distance, Girvan-Newman Benchmark and Normalized Mutual Information algorithms were developed in JavaScript. To visualize the results, an interface using D3.js (SVG and Canvas) and Cytoscape was implemented. Every community finding algorithm was tested in terms of accuracy, speed and memory against 2 synthetic networks (Girvan-Newman and Lacichinetti-Fortunato-Radicchi networks with varying parameters). Final goal was to cluster microbiological data.

Check out more in the thesis website. You may also download an image of the application in Docker Hub.

Supervision Team

Bruno Gonçalves (iMM) | Alexandre Francisco (INESC-ID & IST) | João Carriço (iMM & IST)

I am deeply grateful for their help along this unique journey...