Getting high precision is difficult when done on "all content", especially considering the multitude of languages and dialects in Europe. At a certain point, more data does not result in better output, sometimes it's actually detrimental

The patterns that arise from metadata are much more generalisable and there's less of it, so it's easier to search through