In the sidebar of HubMed's Tag Storage user pages there's now a new 'recommendations' link. If you have a library of tagged articles, the 'recommendations' function will look at the abstracts of all those articles, select over-represented terms and conduct a query based on those terms, ignoring articles that you've already tagged. The result will be a page of recommended articles that are closely related to, but not in, your original library.
Fetching recommendations from a set of articles in your library that are tagged with a particular tag also works, but it only removes articles that are in that set, so the results may contain articles that are already in your library but don't have that particular tag.
This is similar to the idea described in Russ Altman's 1998 paper "Updating a bibliography using the related articles function within PubMed", except that it doesn't use the related articles function within PubMed.