Sites like Vivisimo (as biometacluster.com) and iBoogie have offered clustering of MEDLINE search results for a while—and do it well. To be honest, I've never found them very useful, but nevertheless thought it would be worth trying out clustering in HubMed as an experiment.
There's now a new icon next to the search description () which will launch the clustering process. It works on the first 200 documents matching the search terms, ordered by relevance, and uses Carrot2 and the Lingo algorithm for clustering. Try it here.
I need to find a better way of displaying the results, but for now each of the clusters links back to a HubMed display of the first 20 documents in that cluster.
I've also made some fixes to the 'expand search terms' function () so that now it actually works at a usable speed.