Hi, I am answering some reviewer comments on a paper I am willing to publish. He inquires on how CorText deals with the clusters’ stability when running the Louvain Algorithm. I know that Louvain is a not deterministic algorithm that detects communities in a bottom-up fashion and also that concerns have been raised on how stable detected communities are. To my understanding, CorText uses a stream of intermediate communities to solve this. Hence, After every clustering step, nodes belonging to the same cluster are reduced to a single node. Relations between nodes within the same cluster are considered as self-relationships, and links to nodes of other clusters connected to the clusters representative. This input is used to run the following clustering level. The procedure is repeated until clusters are stable. Am I correct?

Hi Matias,
there is no cluster stability management system in cortext. As a consequence, two runs of the algorithm can indeed result in different partitions. In my experience, “well structured” networks composed of a few hundred nodes will still results in consistant community structure. As for the process of cluster construction, we use the traditional implementation by Thomas Aynaud which is present in the networkx library and which follows the procedure described by the paper by Blondel et al. (2008)
Vincent D Blondel, Jean-Loup Guillaume, Renaud Lambiotte, Renaud Lefebvre, Journal of Statistical Mechanics: Theory and Experiment 2008(10), P10008

Thanks, Jean-Philippe, to shed some light on this. We are plotting a roughly 300-node network, so your experience matches ours. We have run the script on our data many times and could not tell the difference. Maybe that is why I thought some specific mechanism was in place.

I found your discussion very interesting. I am working on a network with 100.000 nodes and I was wondering if from your experience, you could give me some insights on the stability of the cluster. How does the size of the network impact the stability? Does the stability increase if the number of nodes increases or is it the other way around?

Hi Matias,

there is no cluster stability management system in cortext. As a consequence, two runs of the algorithm can indeed result in different partitions. In my experience, “well structured” networks composed of a few hundred nodes will still results in consistant community structure. As for the process of cluster construction, we use the traditional implementation by Thomas Aynaud which is present in the networkx library and which follows the procedure described by the paper by Blondel et al. (2008)

Vincent D Blondel, Jean-Loup Guillaume, Renaud Lambiotte, Renaud Lefebvre, Journal of Statistical Mechanics: Theory and Experiment 2008(10), P10008

Thanks, Jean-Philippe, to shed some light on this. We are plotting a roughly 300-node network, so your experience matches ours. We have run the script on our data many times and could not tell the difference. Maybe that is why I thought some specific mechanism was in place.

Thanks again!

Hi Matias and Jean-Philippe,

I found your discussion very interesting. I am working on a network with 100.000 nodes and I was wondering if from your experience, you could give me some insights on the stability of the cluster. How does the size of the network impact the stability? Does the stability increase if the number of nodes increases or is it the other way around?

Thank you!