I played with this problem by using the java Neo4J library.

See my blog here.

The problem with pubmed is that the authors don't have a unique ID and there are many authors having the same first & last name.

Here is a relevant web-server http://meshop.oicr.on.ca/meshop/index.html in which one of my friend is doing his PhD research. This site has indexed over 18 million PubMed literatures and has been devised to address the same problem you have mentioned. However, Author-To-Author comparison is still under development.

Pierre is correct - the main obstacle to doing this is author name ambiguity. This is not a solved problem, although there is active research in this area, which may help get you going the right direction: http://onlinelibrary.wiley.com/doi/10.1002/meet.2009.1450460218/full

You may also want to look into methods that include location of corresponding authors, which has led to some pretty (interesting) results: http://olihb.com/2011/01/23/map-of-scientific-collaboration-between-researchers/

Identifying authors is a very difficult and tedious task, and most organizations won't do it. ZFIN.org identifies authors of zebrafish related papers, if the author works primarily in zebrafish. If you wanted to do a research project, they might make their dataset available, even if it doesn't reflect all authors.

Harvard Profiles provides a disambiguation service. See details here

http://profiles.catalyst.harvard.edu/docs/ProfilesRNS_DisambiguationEngine.pdf

You might also be intereseted in a new(ish) effort, called ORCID (Open Researcher and Contributor ID). "ORCID aims to solve the author/contributor name ambiguity problem in scholarly communications by creating a central registry of unique identifiers for individual researchers ..."