Sandbox
If you type a placename
How it works
To explain what’s going on here, we need to go back a few years to 2005, when Yahoo! acquired a London-based company called Where On Earth. The team from Where On Earth came to talk to us at Multimap about geocoding in the same year; they had a database product and we talked about how/if it could help Multimap get better at finding places by name. They were knowledgeable and lucid and seemed to have a clear understanding of the challenges of placename search (as opposed to address-level geocoding; I don’t think they were ever in that game). Their database was built up over years and de-duping of non-unique placenames was done manually; this is its essential added value. Some spatial relationships are captured in the database schema, particularly ‘contains’ and ‘within’ relationships, e.g. Maida Vale is in London which is in England which is in UK. The relative importance of places is also stored, so a search for Boston will return multiple results, with the result for ‘Boston, Massachusetts, USA’ promoted above ‘Boston, Lincolnshire, UK’.
We liked the product but didn’t take the discussions very far because of Multimap’s other priorities around that time. We wished them well when Yahoo! bought the company, and wondered whether the database would disappear from public view. Thankfully it didn’t; it’s now available to developers via the Yahoo! and Flickr APIs.
Now, whenever someone uploads a geotagged photo, Flickr attaches up to 6 Where On Earth IDs (WOEID) to it (e.g. one each for London, Ontario, Canada). I’m not sure to what extent this is done via point-in-polygon queries, and to what extent via a relational placename hierarchy stored in the data. For a few months (since at least August 2008) users have been able to correct the WOE placenames attached to their geotagged photos, using a list of nearby neighbourhoods, so Flickr is gathering data on their users’ understanding of neighbourhood boundaries. They recognise that it’s a job that will never be finished, though.
The team at Flickr decided to work backwards, starting with the millions of longitude/latitude coordinate pairs belonging to geotagged photos, to generate shapes that t
So the polygons this application displays reflect all these factors:
Flickr users’ ideas of neighbourhood boundaries
the algorithm that Flickr uses to attach WOEIDs to coordinates.
whether there are enough geotagged photos nearby (not sure how many are needed) for Flickr to build the shapes
I think the algorithm for attaching WOEIDs to coordinates allows for some overlap; i.e. a coordinate pair can be associated with more than one named place at the same level in the hierarchy