[okfn-discuss] Help needed with visualization

[Peter Murray-Rust]

This is very exciting and something I have been scratching at in science
publishing. Now the Openbiblio team has produced Bibsoup/bibserver and it
seems your application could be very well suited to for BibSoup

On Tue, Jan 24, 2012 at 4:25 PM, Guo Xu <digitalepourpre at gmail.com> wrote:

> Hi folks,
>
> I have been working on visualizing the networks of academic publishing
> in economics. Here's an example for the Quarterly Journal of
> Economics:
>
> http://www.guoxu.org/econmap/map.html
>
> A link indicates that two economists have published together in the
> QJE. The strength of a link is defined by how many times they have
> published together.
>
> The size of the node indicates how many times an author has published
> in the QJE. Bigger nodes have published more often.
>
> Finally, the color indicates the ranking of the economist's alma
> mater. Blue indicates that the author obtained his/her PhD from a top
> 10 university (according to
>
> http://www.topuniversities.com/university-rankings/world-university-rankings/2011/subject-rankings/social-sciences/economics
> );
> orange indicates a top 11-20 university; green is for top 21-30 and
> red is for all universities beyond top 30.
>
> Couple of interesting points:
>
> - It seems that the core (those at the centre) are almost all made up
> by top 10 authors. They tend to be well-connected.
>

In the UK this might be called "the old boy network" - the unofficial
network of (men) who have been to the same school / university. It does not
necessarily indicate  absolute vaue but it is often correlated with getting
grants, etc. [I have been in both Blue and Red universities (in science)]

>
> - The hubs are: Phillipe Aghion, Daron Acemoglu, Marianne Bertrand
>
> - There are rarely authors beyond the top 30 who get published in the QJE.
>
> The visualization is done with D3. But it is very slow on older
> computers. Does anyone have ideas for optimizing this?
>

Yes. This is a dynamics exercise and (I assume) you have a pairwise
repulsion term to spread the points out. Many of your points are
0-connected and so you spend a lot of time computing them for nothing.
Unless there is some other hidden coordinate  I would just separate into
the disjoint graphs. It will be hugely fast as instead of O[N*N] you have
O[N] or less (there is a power law distrinution of cluster size)

>
> Also, I have a lot more characteristics lying around that can be
> displayed (e.g. gender - btw only 10% of the authors are female), but
> I do not really know how to do it dynamically.
>
> Finally, I would ideally like to do the same visualization for the
> *entire* network of economist. I have a 300 MB dataset scraped from
> Repec that gives me information on co-authoring for virtually all
> economics journals and working paper series. But obviously this will
> be too slow to visualize so it would be great if someone had
> experience in working with such big datasets (the whole dataset has
> ~30.000 economists, which results in a 30.000 x 30.000 data matrix!!)
>
> You will certainly find interest on openbiblio-dev as we are looking for
bibliographic data sets and things to do with them


> Anyway, let me know what you think and looking forward to suggestions!
>
> Guo
>
> _______________________________________________
> okfn-discuss mailing list
> okfn-discuss at lists.okfn.org
> http://lists.okfn.org/mailman/listinfo/okfn-discuss
>



-- 
Peter Murray-Rust
Reader in Molecular Informatics
Unilever Centre, Dep. Of Chemistry
University of Cambridge
CB2 1EW, UK
+44-1223-763069
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.okfn.org/pipermail/okfn-discuss/attachments/20120124/f111101f/attachment.html>