In short, I wouldn't limit myself to Twitter, nor its API, nor tags or structured content, Nova, I would ETL all textual ressources permanently identified and locatable on the whole Web (like tweets, and for longer texts, just machine summarize back to 140 chars). The qualified resources would require some mandatory metadata (location, authorship, for instance).

ETL as in extract, transform and load, of course. You have something for search engines, if I recall correctly? This is my shot at your lazyweb request. Because you have to believe, at least a little bit, that flattering works for me… ;)