Automatic Keyword Extraction from Text (blog.urx.com)

If you want to make it across the "commerialization valley of death" you are going to have to sweep the floor with Alchemy API rather than simply "perform better in some circumstances"

The code's available too: https://github.com/lvsh/keywordfinder

Have you considered using PLSA (Probabilistic latent semantic analysis)? Effectively, this method does very similiar thing as you aim to do - it chooses a set of words that represen a topic of a document.