There's a bug in your tag cloud: You're splitting on whitespace. The tag "venture capitalists", for example, really isn't the same as the tag "venture" plus the tag "capitalists"...
Have you tried automatically learning a vocabulary based on seeing which words go together often? I'm thinking e.g., if the word "Block" is followed by the word "Store" more than X% of the time, there should probably be an atomic "Block Store" tag.
Hmm, this sounds like a fun puzzle. Can you send me your dataset?
