Most common failed queries

[Cj-Malone]

Is there already/has anyone already thought about there being stats on "failed" queries? I think it could be really useful for driving edits in OSM, that effect the most people. Eg if everyone searches for "AH" instead of "Albert Heijn" we'd know to add it as a short_name in OSM, or an alias elsewhere. Or a high amount of queries are for "x Regent Street, London" but they come back as "road" results, we'd know that area really needs better addresses.

[otbutz]

I have mixed feelings about this. While I'm all for improving the OSM dataset, there's a certain risk that this will end up as mapping for the renderer, which is frowned upon.

[Cj-Malone]

"Mapping for the render" is about adding bad data to counter bad software. I'm firmly talking about accurate edits that help users.

[lonvia]

This sounds good in theory but doesn't work in practice because we usually don't know what the actual intent of the failed query was. Point in case, "AH" might be standing for "Albert Heijn" but the query might come as well from a German looking for the town of Ahaus by its license plate ID.

As a general rule, most of the requests where Nominatim cannot give an answer at all are really garbage: IP addresses, URLS, user supplied location descriptions like "somewhere on earth" (very popular with researchers who try to analyse social media posts) or copy&paste gone wrong.

Another large group are queries with typos and there it is really hard to tell if the typo was in the search query or in the OSM data.

[Cj-Malone]

This sounds good in theory but doesn't work in practice because we usually don't know what the actual intent of the failed query was. Point in case, "AH" might be standing for "Albert Heijn" but the query might come as well from a German looking for the town of Ahaus by its license plate ID.

Ah, context is so important. If people are searching for Ahaus by AH, then again I think it should be added as an alias somewhere, but this isn't know if the query is literally just "AH". I guess Nominatim has no concept of a "session", if someone searches for X, then Y the queries would be linked?

As a general rule, most of the requests where Nominatim cannot give an answer at all are really garbage: IP addresses, URLS, user supplied location descriptions like "somewhere on earth" (very popular with researchers who try to analyse social media posts) or copy&paste gone wrong.

I am assuming when the queries are grouped and sorted by count, any individual garbage query would have low counts compared to the real issues.

Another large group are queries with typos and there it is really hard to tell if the typo was in the search query or in the OSM data.

I did notice a while ago there was a repeated issue with misspelling "Ipiranga" vs "Iprianga", I was able to add a match name to NSI, now the issue is basically fixed. It would have probably been easier for a local person, but when all the social media points to one spelling, it seemed like a good move.