street not found : Rue de la Plaine

[etienneJr]

A cartes.app user noticed an unexpected result while searching Rue de la Plaine, in Paris.

• result n°1 has "de la plaine" so I understand why in comes in first
• but results n°2, 3 and 4 do not have "la plaine", I really don't understand why they come before rue de la Plaine

Do you have an explanation ?
Thanks !

[lonvia]

The results are all in the district "Quartier de la Plaine-de-Monceau". The core issue here is that some POIs get rather high importances. Nominatim doesn't have an issue with that because it gives preference to perfect name matches. But that's not something that is really possible with Photon.

I'm working on a post-reranking of results, which rechecks the results against the original query. This should hopefully make the full match results float to the top.

It might also be necessary to downscale the importances for POIs a bit.

[etienneJr]

Thanks for checking this so quickly!
It's good to hear that you are working on some improvements.
However I'am wondering whether searching in the district field is really relevant?

• in this specific case: I live in Paris suburb, I know this place, but I had never heard the name "Plaine-de-Monceau" before.
• in a more general case: I don't think that anyone in France would use the name of the district when searching for a street or POI. People will use the name of the city, not the district. Is it different in Germany or in other countries ?
  So... would it be possible that I disable on my photon instance this search in the district name ?

[lonvia]

It's what is in the OSM data. There is no point in tweaking the data for one special case (or even for one special country). Search has to work well, no matter what data OSM throws at Photon.

If you want to massage the data nonetheless, use the JSON dumps, remove all data that you consider useless and import the result.

[etienneJr]

It's what is in the OSM data. There is no point in tweaking the data for one special case (or even for one special country). Search has to work well, no matter what data OSM throws at Photon.

Sorry if I wasn't clear, of course I wasn't asking to add an exception for France!

I guess there is a lot of data in OSM that is not indexed by photon, right? The fact is that I wasn't expecting photon to index the district when indexing streets and POIs. If you tell me that the district is commonly used in at least one country to search for a place, then of course photon should take it into account. But I was thinking: if no one in the world is used to using district to search for a place, and if you can't manage to post-filter the results to avoid this issue, then maybe it would be easier for you if photon does not index the district of streets, but only their cities.

If you want to massage the data nonetheless, use the JSON dumps, remove all data that you consider useless and import the result.

Thanks for the tip, I'll keep it in mind, but for now I think I'll just wait for the improvements you're working on. Thanks!

[laem]

I don't think that anyone in France would use the name of the district when searching for a street or POI.

I'm not sure about that : this district may not be so well known, but some others are. E.g. Montmartre, lots of people will search this and expect matches on the district. Other examples in Rennes are Thabor, Saint-Hélier, well known neighborhoods that will be included in search strings I guess.

E.g. follow this link (zoom on Rennes), search for "hibou", a well-known bar. No result, everything's really far. Dissapointing. Add "Saint-Hélier" or add "californie" (two names for this place) -> good result.

Just to be sure, I also tried "Hibou arizona" -> doen't work, ouf :D

[lonvia]

The rescoring from #1041 brings the street result into second place.

There is still an issue with overly high importance values on universities and large companies which needs to be solved to get rid of the university in the top position.