Hacker News Reader

Comment by solomonb | Hacker News Reader

Music recommendation is such a hard problem. There are all these seemingly obvious relationships you can map between bands to create a big graph that looks good but that almost never captures what goes on when a human with deep music knowledge recommends music. Often the best recommendations have no obvious relationships to the band you like.

I played around with this tool a bit and didn't find it any better then other more traditional music discovery tools, that is to say not very effective.

For example, I entered John Zorn and was recommended a bunch of John Zorn's bands. I entered The Residents and got The Pixies :/

I think its more effective to click around on Music Brainz and Wikipedia.

View on HN

For me music discovery is a solved problem. Here's the algorithm:

1) Imagine the timeline of musical history. If you don't have a clear idea of it, Wikipedia is a good place to start.

2) Pick any genre/period you don't know well. (For example, medieval music, or swing-era jazz.)

3) Look up the main figures of that genre/period. (For example, Guillaume de Machaut, or Duke Ellington.) Wikipedia is good for this too.

4) Listen to a sample of their most well known pieces. YouTube is good for this.

5) Repeat. Go down rabbit holes when you like.

No fancy tools needed, just your mind and the internet. This will give you interesting music for many years, and improve your musical taste a lot too.

I agree this works really well and do it, this is essentially what I meant when i said 'clicking around music brainz and wikipedia.' That said I wouldn't be satisfied with this as the only way i could discover new music. There are so many dimensions that don't get codified in wikipedia or music brainz.

Good advice.

However for some genres that approach won't work, since they are either too new, too niche, the genre-description says too little about the actual songs etc. If this is the case another tip is to go at it from the production/distribution/scene side. So you check music mixed by the same audio engineer, released on the same record label, made in the same city during the same time. This can get you surprisingly far.

There is no real shortcut to doing it yourself, part of appreciating that music is often also to understand the context within which it was made.

Great suggestion. Likewise exploring who played shows with who is another great way relation for music discovery. Often you can find radically different bands that were part of the same social scene but which you can relate to.

There are so many ways music can be connected that aren't accounted for by genre labels or "sounds like."

You seem knowledgeable about this.. Care to test my old project for music recommendation? I built it by looking at co-occurrence of artists in Spotify playlists, which gives me word2vec-style vectors, and then its just kNN.

No login needed, just enter some artist names and see what you get:

https://blog.jonas-klesen.de/artist2vec

Very interesting, I've been working on a similar project (using word2vec to learn vectors using playlist data), but using songs instead of artists as the 'words'.

The main bottleneck at this point is the volume of data - many songs I'm interested in only are only represented in a handful of playlists, and . Evaluation at any useful scale is also quite difficult. For somewhat obvious reasons, in our AI era Spotify has become quite skittish about letting third parties gain access to their data at scale...

This is pretty neat, shows good relationships especially on the edgecases where an artist has a very unique sound that other artists dont mimic, but otherwise people who typically like that artist will like others.

Would be very cool if it supported smaller artists than it currently does, because imo thats how you start surfacing emerging talent.

Pandora solved this problem nearly 20 years ago an Spotify with all its money and engineers do such a bad job, it’s beyond absurd.

Nothing beats humans with great music tastes and deep knowledge. I’ve yet to find any form of recommendation engine that has surprised and delighted me the way humans have.

This tool might unearth something interesting, but I find it sus that it’s recommended the same artist (Adrianne Lenker) when I asked about Aimee Mann and Steven Jessie Bernstein.

Reading your comment and all its subtree made me realize there's another difficulty to the problem: what atomic unit do you use? Tracks, albums, artists?

One might argue that "artist" isn't granular enough, since lots of (most?) artists change sound during their career. For the two others, I think recommendations should be trained and given separately (segregated, if you will) between people who listen to albums and those who only care about tracks/singles.

the problem is there's different ways that people engage with music. Some listen to the lyrics and want to have an emotional connection, some view it as exploratory art, others wear it as an identity, some are just looking for similar sounds ... You need to have a routing system that can match the recommender to the style of engagement.

If you don't have that, you can't solve it.

If you're into John Zorn and The Residents, you gotta check out Angine de Poitrine: https://www.youtube.com/watch?v=0Ssi-9wS1so

Microtonal polyrhythmic looping absolute madness. (you can hear some Primus and King Gizzard and the Lizard Wizard kinda sounds in there, if they also tickle your fancy)

Residents -> Pixies is certainly an odd recommendation. Having said that, where _can_ you go from The Residents? Daniel Johnston?

> Residents -> Pixies is certainly an odd recommendation. Having said that, where _can_ you go from The Residents? Daniel Johnston?

I would be truly impressed if a recommendation engine took me from The Residents to Balinese Gamelan. My aunt plays in a Gamelan orchestra with one member of the Residents and learning that somehow made so much sense to me. This are the kind of out of pocket recommendation that an engine will never capture.

Spotify seems to have mastered music recommendation.

It would be great if somebody could reverse engineer their recommendation algo

Actually, I find Spotify horrible for finding new music outside my bubble that i like. YouTube works much, MUCH better for me for this purpose.

Interesting. Spotify works almost perfectly for my discovery needs. I just pick a track I know that fits my mood, then use the (3-dot menu) "Go to Radio" option, which leads to a playlist that usually includes tracks and/or artists new to me. It's been a reliable discovery mechanism for me for many years. Also, there's a new feature I first saw within the last week, a "non-personalized" option that increases the "new to me" ratio.

the "you might also like" for a given artist is usually the most generic related artists - for anything remotely related you'll get basically the same list which is the middle of the venn diagram of everyone who listens to them

Hmm, just tried the non-personalized option for the first time. Is it a reflection on me that I may prefer it over "personalized"?

I always find this interesting… Spotify is phenomenal for me - about every third Monday Discovery playlist has two or three hits, which feels like a pretty solid ratio, at this point. YouTube has never suggested a single thing I cared for.

I wonder if it’s a curation thing? I’ve been with Spotify since the first day it was available, and rarely use YouTube. I haven’t had a good music ratio as good since newsgroups and (real) forums a decade ago, which were a different form of curation.

Reply on news.ycombinator.com