You reminded me of a failed experiment I did a while back with music tags. I still purchase music monthly (Bandcamp when possible), and I always tag music. I have a bunch of multi-value tags including 'instruments' and 'moods'. It works really well with a huge collection. I can quickly find songs similar to the one I'm listening to, or generate playlists.

Anyway, a few years ago I thought I could speed up the tagging process by pulling tags from Last FM and filtering out the shite by only including tags that matched an instrument name or mood. Luckily I backed up before running the script on my whole collection. The data was absolute garbage.