Maybe it would be better if we try a more top down approach generating data semantically?

this perspective is 2 years older than this post... even RDF people agree that the semantic web lower case is taking hold

no mechanism that allows or even requires content authors to create their own metadata will ever take off. lets go back in time to 1996 and look at the meta tag. the idea was that people would describe pages with terms like "cancer research" or "fly fishing" but instead used terms like "pussy cunt fuck cum orgy". why? because people figured out very quickly what people were looking for on the web and crafted meta tag keyphrases apprpriately

the whole point of quality search engines like google is to circumvent this garbage....i.e. circumvent the semantic web

microformats just moves the problem into a new area of the text. another flop.

SemWeb and Facebook

Say you want to list name:knows:name, would you rather:

a. use /friends.rdf

b. use /GetFriendsList.php and get a JSON list of friends