[OSM-talk] Nominatim (was Re: Nominatim & US places)

[Steve Doerr]

On 30/12/2010 22:09, Craig Wallace wrote:

> I'm not involved in Nominatim, but some comments anyway:
> You can go to this page and search for a place:
> http://nominatim.openstreetmap.org/
> In the results, click the little link for "Details", which will give
> more details of how Nominatim figured out the address. So for Biitig
> Road it gives this page:
> http://open.mapquestapi.com/nominatim/v1/details.php?place_id=14428803
> As you can see from that page, it is using the place=county node for
> Saratoga. If counties etc are just mapped as nodes, then Nominatim has
> to guess which county a place is in, based on which node is nearest. And
> in this example, the node for Saratoga county is nearer to Biitig Road
> than the node for Rensselaer county.
>
> So its much more useful to map counties as areas/multipolygon relations,
> and delete the place=county nodes. If you have now done this, and its a
> complete border, and tagged correctly, then Nominatim should use it for
> the address.
> The same applies for cities/towns/villages etc - if they have defined
> boundaries, then its a good idea to map them instead of just a node.
>
> Have you got a link for the way/relation for Rensselaer County? I could
> have a look at it, and see if its tagged correctly and complete.

I've noticed a similar problem with my home county of Kent, United 
Kingdom. My own street, Mulberry Road in Northfleet, doesn't show up in 
a search on "Mulberry Road, Kent" (it goes straight to another street of 
the same name in the county). The entry for my street 
(http://open.mapquestapi.com/nominatim/v1/details.php?place_id=13865667) 
doesn't mention any county in fact.

A search for "Kent, UK" brings up this: 
http://open.mapquestapi.com/nominatim/v1/details.php?place_id=251089, 
which is based on a single node and very inaccurate (and, not 
surprisingly, my Mulberry Road is outside that boundary).

And yet there is a relation for the county of Kent: 
http://www.openstreetmap.org/browse/relation/172385

Is Nominatim ignoring this relation for some reason?

-- 
Steve