QRank

[1ec5]

I recently came across QRank, which provides regularly updated ranks for Wikidata items that are designed for map-related applications. Maybe it would be useful signal for ranking search results, though I don’t know to what extent it overlaps with the heuristics already implemented here.

[lonvia]

Interesting. QRank uses a different metrics (page views) then this project (page rank). So it would be interesting to compare how the two metrics do. Or try out a combination of the metrics.

[joelmellon]

I think this is a really great idea.

I just wanted to point out that it's a relatively easy thing to incorporate into projects. Assuming you have a place's Wikidata Item ID (eg. you're using Nominatim or other OSM data) you can easily derive its QRank. Since the generated (and occasionally published) QRank data file is simply in CSV format <item-id>,<rank> getting a place's rank is a simple look up, eg. a single database join.

The only tricky part in my mind is rescaling QRank to match the existing implementation of importance here. Is importance just a multiplier? Is it scaled to the min and max results? etc. Maybe treating it like a replacement for the current implementation isn't the right way of thinking though. I can imagine reasons to use both, or for one to augment the other.

[1ec5]

Another consideration is that the page views for a given item may be due to something other than the subject’s inherent geographical importance. qrank-map visualizes the impact of weighting QRank heavily as of December 2022. For example, two of the 23 features in the contiguous United States at zoom level 3 are Google’s Mountain View and New York City offices, three are monuments, one is a tree, and one is a city primarily known for a natural gas pipeline explosion. The pattern continues as you zoom in, prioritizing the headquarters of major tech companies and banks, sites of mass shootings, and mothballed airplanes on display at museums. I can’t explain the airplanes, but the other topics have been breaking news events at some point or another, driving massive amounts of traffic to Wikipedia.

The good news is that a search form factor tends to obscure surprises like this somewhat better than a rendered map would, but it probably shows that QRank could only be a small factor compared to what Nominatim is already using.

[joelmellon]

Wow, great insight @1ec5 !