Community Discovery Network

Download the Algorithm Similarity Network

This is the companion supplementary material for the paper “Discovering Communities of Community Discovery”, currently under review to the ASONAM conference. The package includes the data & code to replicate the main results of the paper.

If you’re only interested in the main Algorithm Similarity Network, you can find it in the asn.tsv file in the main folder of the ZIP archive. The network’s nodes are algorithm solving the community discovery problem. I ran more than a thousand benchmarks and I counted the number of times in which the two algorithms returned very similar results, which is the basis for the edges. This version of the network has been filtered using the noise corrected backboning technique. The file has four columns: the first two are the algorithms’ names, and the second two are: nij (the number of benchmark in which the two algorithms agree) and score (a measure of significance of the edge weight).

Otherwise, the replication material is divided into two folders.

01_network contains the full non-backboned network (asn_all_complete_top5count.tsv) and various view of it, plus Python scripts to generate figures 1 & 2 in the paper.

02_robustness contains several alternative ways to build ASN: by averaging similarity scores (asn_all_complete_nmiavg.tsv); imposing a hard threshold to count the similarities between two algorithms (asn_all_backbone_hardthresh.tsv); only considering LFR benchmarks (asn_lfr_complete_top5count.tsv); or only considering real world networks as benchmarks (asn_real_complete_top5count.tsv).

Note that the Python scripts require a set of standard Python scientific libraries (numpy, scipy, pandas, scikit-learn, networkx, …). The package also includes compiled binaries to calculate Infomap communities (code from Daniel Edler & Martin Rosvall) and the overlapping normalized mutual information (code from McDaid’s GitHub). I compiled them on Ubuntu 18.04, thus they might not work on other systems and you’re encouraged to re-compile them yourself.

Download the Algorithm Similarity Network