Different Graph Metrics results

Sep 9, 2010 at 6:46 PM
Edited Sep 9, 2010 at 7:04 PM

I am now learning Nodexl by going thru the book: Analyzing Social Media Networks with NodeXL Insights from a Connected World. I am using Version 1.0.1.131 of the software.

When I opened the kite example data (downloaded from http://casci.umd.edu/images/d/d3/Kite_Example.xlsx), and clicked "Graph Metrics," I get the results attached below (the first image). All of the Graph Metrics results are different from the ones provided in the book (p. 71, the second image), except degree and clustering coefficient columns.

Which are the right results? And what was wrong?

Thanks.

Bryan

[My results]

[Textbook results]

 

Sep 9, 2010 at 8:48 PM
Bryan,

Where did you find a copy of Analyzing Social Media Networks with NodeXL
Insights from a Connected World? I have it on pre-order with amazon.com
and they're saying it will ship sometime between September 13 and October
1. If I could get it sooner, I surely would.

Thanks.

Peter

On Thu, September 9, 2010 2:46 pm, boysh wrote:
> From: boysh
>
>
> I am now learning Nodexl by going thru the book: Analyzing Social Media
> Networks with NodeXL Insights from a Connected World. I am using Version
> 1.0.1.131.When I opened the kite example data (downloaded from
> http://casci.umd.edu/images/d/d3/Kite_Example.xlsx), and clicked "Graph
> Metrics," I get the results attached below. All of the Graph Metrics
> results are different from the one provided in the book (p. 71), except
> degree and clustering coefficient.Which are the right results? And what
> was wrong?Thanks.Bryan
>
>
Sep 9, 2010 at 9:23 PM

Bryan:

Your images didn't make it through, but I think I can explain what you're seeing.

For many graph metrics, there is no single "correct" way to calculate them, and we changed some of our calculation techniques after the book version of NodeXL (1.0.1.113) was released.  For example, we used to normalize betweenness centralities, so they were always in a range from 0 to 1.  The unnormalized values actually have more meaning for some people, so we no longer normalize them and they can now be any non-negative value.

I recommend that you continue to use the latest version of NodeXL, even if it doesn't look exactly like the screenshots in the book.  NodeXL is under continuous development, and we release new features and bug fixes approximately every two weeks.  It's unfortunate that the book is already somewhat out of date, but that's an inevitable result of the slower book-publishing cycle.

-- Tony

Sep 9, 2010 at 10:06 PM
Thanks so much, Tony, for the response. The betweenness centrality makes sense now in light of the new results; the values I have range from 0 to 14. I remenber the book also mentioned you reversed the closeness centrality scale sometime down the road of developing the software. Are the formulae or the ways in which the graph metrics are calculated (betweenness centrality, closeness centrality, eigenvector centrality, etc.) associated with each released Nodexl version (particularly the most recent one) publicly available on the Internet? I don't know what happened to the closeness centrality and eigenvector centrality calculations. I may need to know how they are calculated to make sense of the data and to report them in papers anyway. Thanks a lot. Bryan
Sep 10, 2010 at 4:12 PM
Edited Sep 10, 2010 at 4:13 PM

Bryan:

If the "Details" links in NodeXL's Graph Metrics dialog boxes don't answer your questions, let me know what they are and I'll try to answer them.  I can also point you to the source code if you are interested in the nitty-gritty implementation details.

For closeness, betweenness, and eigenvector centralities, and also PageRank, NodeXL uses a code library called SNAP (Stanford Network Analysis Package), developed by Jure Leskovec at Stanford.  Questions about those particular metrics are best answered by Jure.  You can find his contact information at http://snap.stanford.edu/about.html.

-- Tony