I've taken a look at recent posts on locations and geo-tagging. My
read of all of this is that we can associate tweets with locations in
1. Geo-tags (user opt-in)
2. Location (user provided, pretty um.. low quality)
3. Some kind of behind the scenes magic that Twitter is doing
For case 3, that means that when we specify geo-boxes we're getting
something more than just 1. Is there anything available publicly about
how this is done? i.e. is it parsing of User.location, some kind of IP
thing, spy satellites..? ;)
As someone posted a while back, it seems that we can get all tweets
within a geo-box, but we can't get the inverse, i.e. an (approximate)
lat lon for an arbitrary tweet. So suppose:
Tg = all tweets within box g.
Tg[k] = some tweet in that bounding box, *without* a geo-location
>From which I "know" that Tg[k] is in g.
Now, is that based solely on info from U[l] or does it take into
account anything about Tg[k]?
And, is my understanding correct that if I discovered Tg[k] from
somewhere outside of that location search, I *can't* determine g
(unless of course it is geo-tagged or I do some kind of bone-headed
exhaustive search..) ?
Finally, has anyone else in API-consumer land come up with a good set
of heuristics for determining location from the user.location alone? I
mean, there are some obvious steps, but I don't want to re-invent the
wheel and given the uncertainty about the data available
("TeaPartyVille,USA", "Beer City In Flavor Country" (sounds like a
nice place to visit)) I'm not certain it's worth it. Are people having
pretty good results about just parsing place names? Code? :)
And of course, does anyone want to tell us/speculate about what
TrendsMaps is doing here? My assumption is that they are just doing
searches based on the twitter geo-boxing, but perhaps there is more
magic here that might be sharable.