[OSM-talk] Name finder and home page search working again

[David Earl]

I'm pleased to say that after a major overhaul of the name finder to
make it possible to update incrementally, and also a move to a different
server (many thanks to Tom for helping set this up), the Name Finder and 
home page search box are now working again with a bang up to date index.

The Name Finder home page is at
http://gazetteer.openstreetmap.org/namefinder and the search API is at
http://gazetteer.openstreetmap.org/namefinder/search.xml

(I anticipate using the top level of gazetteer.openstreetmap.org for
listings pages in time).

I will update the namefinder plugin for JOSM in due course. In the
meantime, there is a redirect in place on frankieandshadow.com so this
now works once again.

As well as main objective of incremental updates (it will be updated
daily now, overnight in UK time), there are some other improvements:

- searching is more efficient(*), especially for unqualified searches 
with only a few results, though there is some way to go still, and gains 
are undermined by the much larger number of results you tend to get back 
  epsecially for English place names (because of the Tiger import).

- Definite articles(**) are ignored, and "&" and "and" are
equivalent(***), so if someone mapped your favourite pub as "Bull and
Bush" and you search for "The Bull & Bush" it will find it.
Interestingly "The Louvre" will also find "Le Louvre" though this is
more by accident than design.

- Ways which I know represent areas are located by the centre of their
bounding box not a node on the boundary, (so searching for Edgecombe
House as reported by Tom Chance on December 20 now works properly)

- All combinations of apostrophe s, singular and plural should now work
(previously it worked if it was mapped with 's and searched without, but
not vice-versa).

- We now get 100% on the "British Museum Test for public mapping
websites" (though bear in mind "National Gallery" finds those in
Scotland and Ireland as well, Tate Modern catches the one in Liverpool
as well, so qualifying with London is advisable). Unfortunately "disused
station British Museum" comes out ahead of the museum proper in the list
because it is nearer London (****)

David

-----------------------------

(*) simple cases (like just a place name) are especially faster. It
is also faster for contextualised searches (e.g. "Hinton Road,
Fulbourn"), but this can be undermined for some searches because of the
very large number of similar place names in the US - for example if you
search for "Regent Street, Cambridge" there are now scores of places
called Cambridge in the database, and it has to check for Regent Street
in every on of them. So it is not that these are particularly
inefficient searches, just that there are a lots and lots of them to do.
Many place names in the UK suffer from much-replicated US villages of
the same name - so for example, even searches like "Warwick Avenue,
Woodbridge" are quite slow because there are 16 possible Woodbridges to
search - and indeed there are positive results in or near five of them
(though it's still faster than before the TIGER uploads). Suggestions
for addressing this welcome - incrementally returning results is one
possibility I guess. You can qualify it "..., UK" (provided the is_in is
present - which is almost never is for US places), but that's non-obvious.

(**) The, Le, La, Der, Die, Das, El and Il.

(***) Non-English support is limited, but in principle I could extend
this. At present "Arts et Métiers" station in Paris will match exactly;
and were it mapped "Arts & Métiers" would also match itself, but "Arts
et Métiers" would not match "Arts & Métiers" or vice-versa, though
slightly oddly "Arts and Métiers" would match "Arts & Métiers" had it
been mapped like that.

(****) See talk thread "The British Museum Test" starting Nov 24 2007,
and
http://povesham.wordpress.com/2007/11/23/the-british-museum-test-for-public-mapping-websites/
"The Tower of London" fixed by the definite article fix; V&A and British
Airways London Eye fixed on the map; St Paul's Cathedral fixed with the
apostrophe s fix, though actually the mapped name should really have an
apostrophe).