Iterated calculation of node values

May 6, 2011 at 1:49 PM

Hi

I am interested in the construction of relatively simple networks that function as models of social processes (e.g. development aid projects). In particlar, models where the nodes have value that are dependent on the values of other nodes they are connected to. This requires reiterated calculation of node values, and some means of displaying the changing node values over time.

I have managed to construct a crude example, involving 5 actors, whose self esteem score is calculated according to the self-esteem of others they are linked to, weighted by the value of those links (which can be positive or negative). You can see the network diagram here (2nd one down the page) and here is a link to a NodeXL file, with the underlying data. See the Iterations worksheet, as well as Edge and Vertices worksheets

[I have used the model in the linked blog to show how unpredictable the behavior of even a simple complex system can be, when there are sufficient feedback loops]

Question: Is there any way of animating this type of model, to show node values changing over time?

And is there a simpler way of constructing the formula for the node values that need to go in to column AC in the Vertices worksheet? With more complex models (i.e more actor, more links) this could become very difficult to do manually

regards, rick davies

Coordinator
May 6, 2011 at 4:44 PM
Edited May 6, 2011 at 5:12 PM

Rick -

Thank you (again!) for your interest and engagement with NodeXL!

What a great blog post and set of ideas!

As mentioned in another post: (see: http://nodexl.codeplex.com/discussions/256244) - Time is of the essence to the NodeXL team.

In that post I describe some of our team's ideas for a feature that we think of as Automated Dynamic Filters.

That idea overlaps but does not quite get at what you describe in your message and blog post.  

I think you are proposing a feature we might integrate into our Graph Metrics features that will iterate over the graph and build add one or more (potentially thousands?) of "generations" of values as columns in the workbook.

One metaphor would be exposing all the intermediate location values assigned to every vertex by the Fruchterman-Reingold layout algorithm written out as a series of pairs of columns appended to the Vertices worksheet.

Your suggestion seems like it requires two things: a dialog interface in which the user specifies a model and a way to run that model in our existing Graph Metrics framework.

Today we make use of some of our own code to calculate network metrics and rely heavily on the open SNAP library from Stanford (see: http://snap.stanford.edu) for most of the performance intensive metrics.

It would be great to hear more about the nature of the model builder that you imagine.

We would also need to do some work to make additional metrics and user generated models easier for people to select and integrate into their metrics generation.

The result would be a Vertices worksheets with potentially 100s or 1000s of additional columns!  This may pose some UI and performance challenges that require additional work.

That said, I can imagine our model (which has not yet been implemented) for Automated Dynamic Filters would support the iterative display of each state of the network.

Lots of details to work out: we look forward to hearing more from you!

Regards,

Marc