[OSM-talk] Nominatim weakness

[Sarah Hoffmann]

On Tue, Dec 15, 2015 at 02:13:06AM +0000, Dave F. wrote:
> On 14/12/2015 08:25, Sarah Hoffmann wrote:
> >
> >Some helpful person has put a wikipedia link to the Starbucks
> >wikipedia page on every single Starbucks in Japan. That's what
> >throwing off Nominatim. Having a wikipedia page boosts the importance
> >of an object.
> 
> Have you considered that the program is over weighting the
> importance of a wiki page?

Yes.

> Do end users want to find a coffee shop local to them or one
> thousands of kilometres away just because it has an extra tag
> attached?

I sincerely hope not.

Given that we have a simple data issue at hand here and that
the target audience of osm.org are mappers who happen to have
the knowledge and skill to fix data issues, one would hope that
a positive feedback loop unfolds and both the bad data
and the bad search results are gone in no time.

> >  And in this case the boost is quite large because
> >the Starbucks wiipedia page is pretty prominent.
> 
> Prominent to who? Could you expand your explanation please?

The importance of wikipedia pages is computed essentially
via a classic link count (how many pages link to it). A wiki
page that describes a global company is prone to receive
a higher link count than the page for a single POI. In fact
that is entirely intended. After all, the whole point of using
wikipedia links is to figure out how universally known
the place is.

But even without this importance weighing, the mere fact that
an object has a wikipedia page is already a good indicator
that it might have a higher relevance. That's why a wikipedia
tag boosts every result.

Sarah