Reduce importance of non-geographic entities

[lonvia]

I'm increasingly seeing issues with high-ranking wikipedia pages on non-geographic topics, in particular companies or institutions. These tend to get mapped on real-world structures that then outrank better matching results. Examples are branches of McDonalds or Freeport-McMoRan Central Analytical Service Center. It's not so much an issue for Nominatim which can compensate with more elaborate matching. But you see a lot of odd results float up in Photon with its more lax matching.

It looks like we might be able to deduce if a wikipedia page refers to a geographic feature by checking if the coordinate location property in its wikidata page is present. We could either just export its presence as a flag in the wikipediadata CSV and let Nominatim adjust importance or reduce the importance value directly. I'm leaning towards the former because that leaves the option open to use the CSV also for non-geographic tasks.

[lonvia]

And here is an example where a simple coordinate flag won't work: https://www.openstreetmap.org/way/63979460 -> https://www.wikidata.org/wiki/Q1780212 The wikidata entry has a coordinate location even though the campus of the university is spread over three different sites.

Do we need to bring back lat/lon in the CSV and have Nominatim compare location, maybe?

[lonvia]

Here is another idea: how about checking for the presence of a headquarters property? That might work as an indicator that it is an organisation rather than a geographic feature.