Comment by danpalmer
1 month ago
Yeah, so I tried again for Sydney and London, and all of the results are really bad.
"Drivers Triangle" appears to be a "park" so small it's not marked in green on Google Maps. It's also in Penrith which is like a 40 minute drive from the city. Rea Reserve is similar. Astrolabe park is... a park, but it's not in the top 10 parks in the city.
Same for London. Belsize Wood Nature Reserve is a very odd pick a long way on public transport or driving from the centre, and despite living in London for 10 years I've never heard of it. Meanwhile Hyde Park, Regents Park, St James's Park, Battersea, Greenwich, Green Park, ... there are so many iconic parks in London.
My advice would be to curate these per city. It's going to be much easier to just decide which the top parks are for any given city, and with a few hundred cities you'd get pretty good coverage of queries.
You’re not wrong at all. Automated discovery breaks down fast in big global cities because raw tags don’t capture “iconic” very well.
I actually just pushed an update that heavily retunes the scoring, boosting national and royal parks and penalizing tiny or generic parcels, which knocks out a lot of the “Drivers Triangle”–type noise. It’s meaningfully better now, but Sydney and London are good examples of where the limits show.
Treating Hyde Park differently from some obscure patch of trees using only OSM-style data is a genuinely hard problem, and at a certain point manual curation just wins. I’m likely going to add curated overrides for major hubs so the obvious, culturally important parks always surface first.
Appreciate the concrete examples. This is exactly the kind of feedback that helps tighten things up.