Home > Datasets, experiments, what i've been doing > Initial results – HGCE, KCLIQUE, InfoMap and LinkClustering

Initial results – HGCE, KCLIQUE, InfoMap and LinkClustering

We ran each community finding algorithm over the datasets: MIT-NOV, Cambridge, InfoCom 2005, and InfoCom 2006. We used bubbleRAP, BubbleH, Unlimited flood, hold & wait, and Prophet to measure the effect of the CFA on the dataset. The results below are grouped by dataset. Only the best results for BubbleH/BubbleRAP KCLIQUE and HGCE are shown. Note: The local ranking used for InfoMap is based on the rank given by the InfoMap algorithm, NOT centrality as with BubbleRAP and BubbleH.

MIT-NOV

MIT-NOV

CAMBRIDGE

CAMBRIDGE

IncoCom 2005

IncoCom 2005

InfoCom 2005

InfoCom 2005

My next plan is to link statistics about community structure (number of communities, average members etc.) to the results.

Also, I want to find out if there is an existing measure of overlappiness? Does the amount of overlap in communities affect the results in any way? Also, what about the extent to which there is ‘nesting’ of communities, does this make a difference?

I also want to visualise the communities produced in some way, so that it is clearer what the network structure for the best results is based on.

This excel file this excel file contains the combined results of each CFA for all 4 datasets, so we can compare the effect of community structure (number of communities, average member size) directly with results for each simulation run.

  1. No comments yet.
  1. No trackbacks yet.