OSM Namefinder again

Hello!

When I type “Texas, USA” in http://gazetteer.openstreetmap.org/namefinder/ the searching result does not include the real Texas site. But when I do the same thing in Google maps, it works perfectly. For some places with doubts, Google maps will popup up with several suggested options for you to choose. So, I am just wondering whether there is any way to improve the OSM’s namefinder. Or is there anyone there working on that? Or any suggestions, any clue to improve?

By the way, in Google maps you don’t need to type your keywords accurately or completely, Google will match your keywords intelligently. So, I am also thinking that maybe OSM should have a similar function. Any suggestions, or clue to develop this part?

Best
Xintao

Bloom filters are nice if you want to spell checking of results.

This or anyother page about spellchecking, perhaps the ordinary spell checking engines might work for this.
http://www.csc.kth.se/tcs/projects/swedish.html

Sadly the problem is harder because it’s often about getting more accurate results than about getting more results.

OSM Namefinder (http://gazetteer.openstreetmap.org/namefinder/) is dying? I can see on the main page:
Index from OpenStreetMap data up to 9 Jan 2009

And I was wondering why it would not find streets in my town added a week ago! :-))))

Is there some working way to find things in the OSM?

The namefinder has been discussed on SoTM 2009 and this is what I (seem to have) remembered: The old namefinder was not being updated anymore in anticipation of a new one, which did not really exist and was maybe in early development. The result was that a new polish round would be applied to the old namefinder and update it’s database again. Help on developing a new namefinder was offered by Nestoria.

What happened since then AFAIC is that several updates for the old namefinder have been attempted but failed for various reasons. Don’t know about the polishing or new namefinder developments.

In short (IMO): this is one of the big fuckups of the OSM project so far which pisses me off because yournavigation.org heavily relies on namefinder.

</end of personal rant>

The current namefinder is fairly complex and, IMO, tries to do too much. A simpler, regularly updated, search engine restricted to the highway and place keys only would be preferable to an enormously out of date system.

I’ve been meaning to do some experiments towards just such a thing but I’ve been too busy adding data or getting on with other things. :frowning:

A new search has been under development for a while, and you’d be able to try it at http://katie.openstreetmap.org/~twain/ except for the fact it’s currently out of action for a ground-up reindexing! It’s pretty good though, and fast too.

Personally I can’t stand the fact that Google tries to search for stuff I didn’t ask for. If I wanted to search for “Oxfordshire” I’d have typed “Oxfordshire”. I typed “Oxford”, so I don’t. Gah.

Thanks for the link, will take another look soon.

Same here, only with “Gloucestershire” and “Gloucester” respectively. :slight_smile:

Name Finder is a first class feature. Not having an up to date index reduces the interest of making map updates…

For example, I just did a massive update from french cadastre for “Vitrolles” in south of France, which is badly mapped by other mapping service (google, mappy and others). I did so as to let my friend find the places where she had to go for her job in this city. But now, she still can not find the places even if they exist in the openstreetmap database…

I’m eager to help in taking some times to understand why the update went wrong… The namefinder source is available. Is it possible to have some sort of access to the computer where the update was done ? Who can give me some starting point to investigate why the update failed ?

-Mathieu

Did you take a look at the system Brian Quinion has developed, which Richard linked to earlier in this thread?
That’s working very well, although seems to treat suburbs as subordinate to hamlets for some reason.

Yes I tried http://katie.openstreetmap.org/~twain/ , but it could not find a single city in france (even Paris, France do not work). I’ve sent a bug report.

I tried to install namefinder on my personnal computer and it seems to work as adevrtised, except that it needs a version of php compiled with large file support to handle the size of nowadays osm databases (several gigas). In fact it ran out of diskspace for the mysql database holding the index, so I didn’t do the full index import (even of only the France part of the database).
But it looks like it would have success on a computer with enough disk space…

I do not know if other limitations exists that prevent the update of the index… I’ve sent a mail to namefinder’s author to ask for some explanations of the problem, but I did not get any response for now…

There is a thread here : http://www.mail-archive.com/talk@openstreetmap.org/msg20303.html which seems to implie that namefinder import is too slow for the size of the actual database.
-Mathieu

PS.For reference, another Thread about Brian’s software : http://lists.openstreetmap.org/pipermail/geocoding/2009-September/000007.html

It doesn’t seem to find anything outside the UK so I assume that Brian only imported the UK. I did think he was importing a full planet. Ah well, sorry for pointing you in the wrong direction!

The new name finder ( nominatim http://wiki.openstreetmap.org/wiki/Nominatim)) has just been updated to the full planet and is now live in the search on the main site. It should now find streets and cities world wide and it should be kept fairly up to date. It also supports house number search and tries to specify more accurately which cities a road belongs to by using administrative boundary polygons were they exist in the data.

Give it a try again and see if it works for you. I have had pretty good results so far and I think it is a big step forward.

Thank you. It indeed works better.

It seems I’ll have to update the location with the is_in tag to have a functionnal search…
It finds “Les plantiers, vitrolles” but it doesn’t seem to link it with “vitrolles”, and it doesn’t find “Le Village, vitrolles”. Where “Les plantiers” seems to be a uniqe name in the database but “Le Village” is quite common… I’ll try with the “is_in” tag… Are also street expected to bear the “is_in” tag ?

So what about an entry in the news-section of the OSM-wiki?

And is there already an announcement in the relevant OSM-mailing-lists?

I don’t think is_in is necessary or desirable. Particularly not on all streets or all housenumbers. Defining a polygon with the appropriate admin_level boarder should make the new search handle the roads correctly.
If you go to nominatim.openstreetmap.org and search through that, there is a “details” link on the result, that will tell you how the result was constructed and how it links street, village, region and country. I.e. if it used nearest neighbour search, or had a proper polygon defining the feature.

I don’t know if Le Village might be a problem, as it could be filtered out as a stop word as it is too common. But I don’t know how it works so not sure about that. You could probably report the issue through the “Report a Problem” button on the nominatim page.

As it has only just gone live, I guess there are still some teathing issues that the “beta testing phase” didn’t catch but twain47 has been very busy trying to fix them all as the reports come in. This is also the reason it hasn’t been announced on wiki news or the mailinglist quite yet, to not overwhelm him with the potential rush of new problem reports. Once everything settles down, I am sure it will get announced appropriately.

Thank you for your detail response…

Doing some research by myself, I’ve figured that the boundary relation was indeed used by “nominatim”. In fact “Le Village” seems to not have been in the database at the time of edit… I’m afraid I made a noisy and inappropriate bug report…

I did some more tries, and the tool seems to work very well !
I’m really impressed by the achieved results (the detail link is stunning !)

-Mathieu

YIPEE!! It now finds my street and house number very well! It uses the administrative boundary and nearest hamlet identifier to get the full address, location and city description. WELL DONE!