[OSM-talk] Nominatim on the main page

[Sarah Hoffmann]

On Mon, Feb 19, 2018 at 10:59:31AM +0100, Maarten Deen wrote:
> On 2018-02-19 10:17, Sarah Hoffmann wrote:
> > On Sun, Feb 18, 2018 at 08:12:45PM +0100, Maarten Deen wrote:
> > > On 2018-02-18 20:07, Paul Johnson wrote:
> > > > On Sun, Feb 18, 2018 at 12:53 PM, Maarten Deen <mdeen at xs4all.nl>
> > > > wrote:
> > > >
> > > > > On 2018-02-18 19:28, Tom Hughes wrote:
> > > 
> > > > > I can't comment about how the algorithm works because I don't know
> > > > > anything about it. I'm just saying that we do tell it the viewbox
> > > >
> > > >  It appears to me that the bounding box is used when searching places
> > > > (towns, cities) or streets, but not when searching objects like shops
> > > > or restaurants.
> > > > For instance, searching for a McDonald's always gives me the
> > > > McDonald's at 1351, George Dieter Drive, El Paso City, El Paso County,
> > > > Texas, 79936, Verenigde Staten van Amerika
> > 
> > To fix that please delete all the wikipedia=McDonalds tags from
> > the McDonalds restaurants that show up inappropriately. Nominatim uses
> > the wikipedia links to determine how well known a place might be and
> > ranks places with a wikipedia tag higher.
> 
> I would expect that a bounding box has precedence over other tags. Why would
> a wikipedia tag have precedence over the bounding box and a name tag not?

It is not a question of precedence. Nominatim looks at different
factors at the same time: the view box, how well-known a place
is (aka wikipedia importance), how well the name of the place matches
your query etc. It takes all these into account weighs them against
each other and comes to a ranking of results.

> But then I still don't understand.
> In the bugreport I have the example of the shop "kruidvat" (it is a chain of
> stores in the Netherlands).
> The bounding box is 6.16575,51.36926,6.17049,51.36759 which centers on the
> Kruidvat store in Venlo [1]. Nominatim returns the kruidvat in Amsterdam
> [2].
> Both nodes have a website tag with the same value. Both nodes have the same
> tags, expect that one has a source tag.
> Then still, why is the boundingbox not looked at _at all_? It's not like
> it's the second or third result, it is the 12th result where all other
> results have similar tags and the results are the same whatever bounding box
> you use.

Interesting. So in this case the importance actually happend to accidentally
cancel out the viewbox influence. I've pushed a preliminary fix to the osm.org
instance. It won't fix the McDonalds or Walmart issues though. They are problems
of a different kind.

Sarah