Paperscape maps the dazzling universe of scientific research | Damien George and Rob Knegjens | Science | theguardian.com
referring to
Paperscape, arXiv.org e-Print archive
They did a cluster analysis in an ingenious way. They used some galaxy-formation code, but with papers instead of galaxies on a 2D surface. They made the papers repel each other, but reduced for papers that refer to each other.
They found that the papers fell into well-defined regions, with each specialty's papers being closest to each other. But which other ones they were closest to was also interesting.
Experimental nuclear and high-energy physics formed a big blob, with astrophysics and theoretical high-energy physics near each end. Between the latter two was general relativity - quantum cosmology.
In the other direction from theoretical high-energy physics was mathematical physics, then condensed-matter physics, then quantum physics.
There were a couple others that were rather distant, like mathematics, quantitative biology, and quantitative finance.
It would be interesting to extend this work with other big collections of papers, like at Google Scholar. I don't know how useful collections of abstracts like PubMed and HighWire might be. One might have to look for other sorts of overlap, like in authors and contents of abstracts.