Twitter Search Networks - key to good networks like in the gallary?

Oct 17, 2012 at 8:53 PM

I've been playing around with NodeXL for awhile now, and I really know my way around well.


The only thing I can't figure out is where I'm going wrong with my Twitter search terms. I've tried to recreate the networks Marc Smith just uploaded last night/today, same keywords and all, and all the networks I get are no where near as complex as his. I know Twitter is constantly evolving, and I don't expect the same identical networks, but all I ever seem to get a a bunch of isolates, with a couple edges here and there. Or when I do get more edges, I one big hub, but never multile clusters with edges between them.


So, I guess my question is, now that I've got the tool down, how can I improve my search terms to get the best network?


Thanks for all your help!


Oct 17, 2012 at 9:02 PM

I think your results are a reflection of your data query results.  When you access Twitter under the normal rate limits it is very hard to get enough data fast enough to make enough maps of interest.  Data access is an issue!  All of the social media platforms have restrictions on data access.  Our project complies with their rules.  We deployed NodeXL Graph Gallery so that users who lacked data could share data provided by users who were able to access data.  I imagine the Graph Gallery as acting as the clearing house for many users to share their small snapshots of data.

So, you might consider turning off the follows graph edges in your queries.  This will lead to having a less dense graph, but it will let you build a network much more quickly.  The resulting reply and mention edges are less frequent (it is easy to follow someone, harder to reply to them) so the networks are more sparse but arguably display the more meaningful interaction edges.

Oct 17, 2012 at 9:30 PM

Thanks for the reply Marc.

I think you pointed out where our graphs differ. I rarely include "Follows relationship", much for the reasons you eluded too. And now that you pointed that out, that makes sense that there would be much more defined clusters when including "Follows" than just basing the network off of Replies and Mentions.


Thanks for the help! Now back to scrapping more networks!