failure to get correct street when housenumber is missing

[etienneJr]

Hi, team !
Several cartes.app users reported failure when housenumber is missing, for example here
I noticed it is a reccurrent question here (#765 #773 etc), but this case seems slightly different, since adding in the query the housenumber (which is absent in OSM) prevents photon from finding even the street.

avenue de serane marseille works perfectly, with or without "la", even with a wrong location bias

11 avenue de serane marseille, without location bias, doesn't find the street at all (and it is terrible if you apply a wrong location bias)

11 avenue de serane marseille, with correct location bias, finds the street in 4th position

Is there something we can do to improve this ? (on photon itself, or on cartes.app instance)

It is completely counterintuitive for the average user to think that they have to remove the housenumber to find the street... or having to search for the town first so that the location bias applies before searching for the street.

I noticed that adding a filter by layer=street works, which is great
https://photon.komoot.io/api?q=11%20avenue%20de%20la%20serane%20marseille&layer=street
so a solution could be to notice that the query contains a street keyword to lauch the query with this filter ?

Thanks !

[etienneJr]

some follow-up on this question:

I tried to used the filter layer=street:

• it works perfectly to get the street when the housenumber is not present in OSM
• BUT housenumbers present in OSM are not found any more, the result gives only the street

So I tried the combined filter layer=house&layer=street:

• housenumber present in OSM are correctly found
• BUT when the housenumber is absent, the street is not found any more.

It seems to me that the (bad) results of the layer=house filter take priority over the (good) result of the layer=street filter.

How can I adjust the scores of the different searches to correct this? Thanks !

[lonvia]

This is already tracked in #204. Eventually, Photon needs to be able to handle that automatically. Probably means that we need to do the "term may be missing" lenient search manually instead of relying on OpenSearch's heuristic.

[etienneJr]

Wahoo this issue is 10 years old ! I didn't even know photon was so old ! 😅

When you say "we need to do", who is "we" ? Do you think someone will ever have time and skills to solve this issue ?