Currently it relies on having some known blog sites in the URL. So it misses "krebsonsecurity.com" and "joelonsoftware.com" and "schneier.com". On the plus side, it's not really sure what most of those are, so leaving it for a second level classifier could be cool.