I thought about trying this, but thinking about how to execute it already sounds painful enough.
For input data, I could use my existing library of mostly individually-selected songs, currently at size 1,662. Since I mostly listen to everything, this spans a rather large range of dates.
Then start taking random songs, and rating them on 1 - 10 scale in relation to entire library, enter ratings into 10 year buckets, and use mean of those ratings.
Probably 5 ratings per bucket to keep it short.
Unfortunately, I most likely can't fill every bucket, hell, some would remain empty. After all, classical music makes my library likely start in late 1600s, and end in 2025.
I didn't think about that. Perhaps I could leave it out, and start at, say 1920s, but that would make the data incomplete.

Problem is, I don't have the years for most of them, so that would mean looking up release dates for those individually.

Huh, what if everyone would absolutely love (old) classical music, but we don't see a spike as the graph starts at age of -40?