Hacker News Reader

Comment by jonty | Hacker News Reader

You should try Spotify's "Discover Weekly" personalised autogenerated playlist, it's been built by the Echo Nest team and learns directly from what you listen to.

https://press.spotify.com/uk/2015/07/20/introducing-discover...

View on HN

I found them to be pretty bad. Let's say I listen to genre X exclusively for two weeks, and then for genres Y and Z during week 3. The genres are completely different (lets say X - rap, Y - kpop, Z - just pop music). On week 4, I'd expect to be recommended a mix of X, Y and Z, seems reasonable. However, it's only X because I listened to it a lot a while ago.

So it's pretty good if you are listening to the same music genre, however, it sucks if you like to change the music based on your mood, etc. I also noticed that it doesn't detect the language the music is in so I often get songs with lyrics in languages that I have no clue about.

When was the last time you used it? I've noticed that the discover weekly playlist has consistently featured four or five majorly different genres, all drawn from the music I've tended to listen to lately. It's been good about doing this since I first subscribed back in May.

Some of the songs have absolutely missed the mark, but there have been others I've found that make me go back and listen to the entire album they're from. Definitely have encountered different languages as well.

A week or two ago :) it was pretty in summer/earlier autumn when I was listening to one genre almost exclusively but I've been mixing up lately and my recent preferences got totally ignored.

I listen to multiple, vastly different genres. I find Spotify's Discover playlist may favor one particular genre one week, but it's usually pretty good.

And the behavior you described doesn't sound "pretty bad". If you haven't listened to a particular genre in a few weeks, I think it's reasonable you don't get suggestions for it for the upcoming week of listening. But if you do start listening to that genre again, your next playlist will have it.

The recommendations are usually on point, but to my knowledge, there's no way to say "I don't like this artist, stop recommending it". I have quite a few of those, that I should like based on my tastes but for some reason I don't.

I could try that. But I think I eventually want something more flexible. For example, when listening to a song, I want the system to automatically pick songs that are similar to that song. Or I'd like to pick songs based on instruments used, tempo, origin, language, genre, etc.

It does that as well, and calls it "radios" for a given artist or song.

Spotify does that already too - right click a song and "Start Song Radio". This will automatically pick songs that are similar.

I know it seems like this is what you want, but for me I've never been satisfied with this simplistic approach to discovery. Taste is very complex, and for me saying to someone "I really love xylophone" and them recommending me a lounge jazz tune with xylophone will always fall flat. What most people don't realise is why they love certain instruments or tempos or genres etc. It's usually down to how they were introduced to these, which defines their archetypal music using these instruments for example. It's all contextual.

My context for liking xylophone goes something like this "I love xylophone, because in the late 90s I listened to a lot of Beastie Boys 'Hello Nasty' which has some really nice pieces of music with some great xylophone in them. Then hearing some Mulatu Astatke recently got me interested in xyolophone and ethio jazz in general, combined with a really cute reaction my girlfriend had to one of his songs."

The best recommendation engine I've seen is Spotify Weekly. It's really very good, but make sure you get in there and save a lot of artists and albums you like first

You can do this if you right click and do play song/artist radio.. but in my experience it sucks really bad, and has only gotten worse over time. Use to, it was kinda useful. Now it's complete garbage. I'm not sure what they changed in their algorithm but typically when I do this I get 5 artists that are only marginally related, and then it proceeds to play music from those 5 artists in a loop and the "dislike this" feature seems to do nothing. I've actually disliked a song and then it came on about 20 minutes later.

I have a playlist of a lot of songs I like (except I try to keep it to one or two songs per band).

The radio from that converges to endless Foo Fighters and The Prodigy -- over half the songs!

It's not that I dislike them, but I don't particularly love them either and there is just no diversity on that station.

I understand, suggestions still have issues. I listen to a lot of more obscurely specific genres, and sometimes will get just a couple of repeated songs if I make a radio based on a lesser-known artist.

That said, if you haven't given the Spotify Weekly Recommendations a shot -- you should. Maybe it's because it has more data to work with than song/artist/playlist-specific radios, but for me that feature has been extremely spot-on. In fact, I hadn't used spotify for a while because I kept hearing the same songs, but now I religiously keep on top of my discoveries each week.

I agree on the weekly discoveries, they are great. The radios, at least the playlist radios, are crap.

It seems like the radios point me towards the most common denominator and plays the most mainstream stuff that matches, while the weekly discovery lists come up with some very quirky stuff that I love.

I guess the radios can't have any "memory", it seems to generate close to one song at a time (hence a lot of repetitions), whereas the discovery list can be created 30 songs at a time and seems to go out of its way to find songs by bands that I've never listened to.

Reply on news.ycombinator.com