I remember when Pandora was attempting to compute the "DNA" of a song (assorted classifications like key, tempo, style) and recommend music based on that. You could then get a "station" based on a single song, and fine-tune it by adding more or disliking songs as they came up.

The end result was underwhelming- it never really captured the characteristics that I actually liked about particular songs, and ended up being crappy or so narrowly tuned that it lacked enough variety to be interesting.

The concept is still around, but with less scientific sounding fluff and, I think, more relaxed parameters for recommendations.

>You could then get a "station" based on a single song

Apple Genius did something similar and it mostly worked not badly because it was drawing from songs in your collection already.

You're more likely to like songs in specific genres and time periods and songs that are popular generally. Once you get beyond that, it gets harder. And the situation is probably even harder with video unless you basically watch superhero films.

Did Pandora change at some point?

The last time I used it, I don't recall seeing the "DNA" nonsense, or being able to get a list of characteristics that they used to build a station around... It's possible that the interface simply changed and I didn't look hard enough for it though.

It's still using the same type of algorithm

Pandora is based off of the "Music Genome Project" which was a massive effort that took tons of music efforts to make happen. Experts literally sat down and deconstructed individual songs by hand and gave them tags.

It's truly a historic achievement and it's almost sad how the massive effort mainly resulted in a private database that's primarily used for a failed Spotify competitor

I am 100% with you there. The things that could be done to advance the creation, understanding and appreciation of music from that work is staggering. It's sad that it's stuck in amber.

I’ve found much better results using your lists to find people with some overlap, and then looking at what they have.

Which is basically what HN is.

It depends what the overlap is of course. Something like music probably has a big age component in what people like for example.

But, yes, in general friends with at least reasonably similar preferences to myself are almost certainly a better source of recommendations for video, music, and books than a recommendation engine.

Do recommendation algorithms not take this into account already? Perhaps privacy policies make it harder to automate this effectively.

The famous ones may, but it ends up taking into account "what do we want you to see/listen/look at" much more into account.

The original Netflix prize also, it turned out, wasn't really implemented for a number of reasons. But one of them was apparently that Netflix doesn't necessarily want to give you the best recommendations; it wants you to keep your subscription. There's certainly some overlap between those objectives but they're not the same thing.

They're also incentivized to show you things that cost them "nothing" or "less" than others things - and if they KNOW the things you'd like to watch it's better for them to string those out so you keep subscribed.

The perfect Netflix customer is one always on the cusp of cancelling from lack of use but never actually does ...

In addition to the cost angle, they're also incentivized to push you towards exclusives. Things you can watch on other services (assuming you subscribe and know they're there) are much less of a hook to keep you on Netflix.

I have played ~5 seconds of one random episode from the front page on my Netflix profile since i subscribed around 2016. This was just to verify that it worked.

It still chuckles me up when they send me an e-mail once in a while, about what I might like based my past viewing preferences.

Note: My GF, has a profile that she uses sometimes, but mine haven't been used since the account was created.