It still doesn't sound very clear...

Guesswork on my part, but perhaps Algorithm::Cluster, based on Michiel de Hoon's cluster program is useful?

A commandline --help page is here; and the paper.

If all else fails there is a clustering bibliography as well.