topic Re: Communities option of Network Analysis in SAS VA in SAS Visual Analytics

Communities option of Network Analysis in SAS VA

JianShen — Mon, 08 Jul 2019 11:32:14 GMT

Hi All,

We are currently trying to visualise a network in VA version 8.3.1, within SAS Viya 3.4. The visualisation of the network in general is going well. However, whenever we add communities to the role section, we end up with more communities than expected. For example, the image below shows that we selected 5 communities in the general option. Yet, the map shows 31.

Whenever we increase the number of communities (e.,g., from 5 to 10), the number of visualised communities increases even more.

We looked at the description of the community option, which says the following:

Yet, as the network shows, we end up with more communities rather than less. As such, we were wondering why this is the case. What is the underlying algorithm/calculation that VA uses to determine the number of communities? And how does VA use this input to decide the number of communities?

The data that we use is structured like this:

We followed the data-structure instructions on this page:

https://communities.sas.com/t5/SAS-Visual-Analytics/Network-Diagram-Links/td-p/158142

Additional options we use are:

- Directed network

- Ungrouped network

- Geomap

- Closeness Centrality

We also tested this without using a geomap, but we get similar results in that case as well. We are mystified by the results of the network and we hope someone can enlighten us on this!

Re: Communities option of Network Analysis in SAS VA

Sam_SAS — Tue, 09 Jul 2019 18:28:58 GMT

Hello,

I wanted to let you know that the development team have been looking at this thread and it appears that there is a bug with Communities at 8.3.

I believe that the bug has already been fixed for the next release. I will see if anyone can suggest a workaround for you at 8.3.

If you would like to create a track with SAS technical support, we can verify that the issue is fixed by using your data.

Sam

Re: Communities option of Network Analysis in SAS VA

Sam_SAS — Tue, 09 Jul 2019 19:19:22 GMT

A suggestion from the development team:

They could be viewing disconnected components, and each component is reporting the correct number of communities but then they add up to more than the control would suggest.

Sam, suggest that they should turn off the map background and then replace ‘community’ with ‘disconnected id’. If that yields multiple IDs for the color, then I suspect what they are seeing is what I mentioned.

Let us know what you find out. If you could create a technical support track, that would be very helpful to us in determining if there is an issue and in fixing any issues.

One of our testers reminded me that there was an issue in the way Communities is documented for 8.3. As you cited above, the 8.3 help says that Communities is the number of "local node groupings", but really it is the number of local node groupings within each contiguous cluster of nodes (also known as a "component"). That is, if you have multiple node clusters that are not connected, then your Communities value will generate that number of groupings within each cluster.

So it might very well be that there the feature is working as designed but the help was misleading.

Thanks,

Sam

Re: Communities option of Network Analysis in SAS VA

JianShen — Tue, 16 Jul 2019 07:59:25 GMT

Thank you for your help! There are indeed multiple clusters within the network. We tried the communities option for just one cluster and the number of clusters identified corresponds to the number of clusters selected. We were slightly confused by the community description and were not aware of the fact that communities indicates the number of local node groupings within each cluster of nodes.

However, we would be delighted if someone could help us with one more question. What is the underlying community algorithm by default? Knowing and understanding the underlying algorithm could provide us with better rationale for selecting the number of preferred communities.

Once again, your help is appreciated and we hope to hear from you.

Re: Communities option of Network Analysis in SAS VA

FalkoSchulz — Tue, 16 Jul 2019 08:34:02 GMT

VA utilizes SAS Viya actions behind the scenes. The one responsible for detecting communities is named Hypergroup. The related documentation is here and should contain additional information and examples. This action is also utilized for determining the graph layout for the non-geographical view.

Hope this helps! Regards, Falko