Ok, how about this: scrape Wikipedia in multiple languages, combine the notability data, and assign weights to appropriate locations. For example notability in Japanese Wikipedia would have more weight in locations where they speak Japanese.

Yes, I realize location != language and it would get tricky in locations with multiple languages like Switzerland or countries with big diasporas. The dataset would still be biased as not all languages have Wikipedia.

It would be a small step into the right direction.