Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Building a Music Recommender with Deep Learning

mattmurray.net

Post reply on HN
Print view View original

41–50 of 74 posts

Re: Building a Music Recommender with Deep Learning

#41

Post by wellboy »

halflings wrote: post #34
wellboy wrote: post #19

Why trying to do it via A.I.? Why not checking which are the top 3 most played songs by other users who are the 1000 users who have the most similarity with the current user, and then recommend the current user the most played songs from the 1000 similar users that the current user has not listened to yet. As far as I can see this would be superior to any existing A.I. recommendation algorithm.

What you're describing is also an "A.I". It's called collaborative filtering, and your algorithm (picking top 3 of the 1000 most similar users) would give results heavily biased towards popular songs, there are better approaches in that field.

Yes, all algorithms are A.I. in that case.

My 1 min effort description would be biased towards popular songs, but you can easily change that by selecting songs that are not popular, but that occupy a lot of playtime with a user.

Re: Building a Music Recommender with Deep Learning

#42

Post by rasjani »

bsenftner wrote: post #30
johnlbevan2 wrote: post #12

Warning: this comment has little to do with the article, beyond being a rant on the approach taken by all recommendation engines I've seen. This an interesting approach, but the objective is similar to most recommendation engines: "Find me something similar to something I like". Sometimes that's a good requirement (e.g. when trying to queue up the next song in a playlist, it's good to have some similarity to the song…

My problem with all music recommendation engines, and for many intellectual music aficionados, the lyrics content - what is being verbally described in the music - is what I seek and hang on for my preferred music. When I listen to my collection, the genres are all over and I don't even know them. I listen to the words and treat the music as emphasis for the words. I'll have ska, 30's jazz, hip hip, and classic rock…

Having a selection that has lyrical continuation from one song into an another is very typical also in reggae.

Reggae as "genre" itself is also quite varied in what goes under its label. There are also other factors that play a big weight on how good matches they are to reference material. Producer and decade make a huge difference but also what's known as "riddim" name should give clues.

Re: Building a Music Recommender with Deep Learning

#43

Post by make3 »

That's not how you build a recommendation engine... You build a recommendation engine by creating an embedding from each song from which user prefers them, as you would for words in word to vec. This is how Amazon and Youtube do it.

https://static.googleusercontent.com/media/research.google.c...

Re: Building a Music Recommender with Deep Learning

#44

Post by pfd1986 »

mattmurray wrote: post #10

Wow thanks for sharing + reading my blog post! I did this for my final project on the Data Science bootcamp at Metis [1] this spring. [1] https://www.thisismetis.com/

Hi. Nice job there. Like others, I was interested in the network architecture. Is the code open source / available somewhere?

Re: Building a Music Recommender with Deep Learning

#45

Post by krosaen »

make3 wrote: post #43

That's not how you build a recommendation engine... You build a recommendation engine by creating an embedding from each song from which user prefers them, as you would for words in word to vec. This is how Amazon and Youtube do it. https://static.googleusercontent.com/media/research.google.c...

Couldn't you view the output of the last layer of the convnet used as the embedding in this case? Yes, this was a different approach than leveraging user preferences, but I don't see why this is inherently the wrong approach.

Re: Building a Music Recommender with Deep Learning

#46

Post by computerex »

dnadler wrote: post #20

Very cool! One minor nitpick -- the author mentions that this is 'completely unsupervised'. It's true that the author didn't need to manually classify the data, but someone did. So, I believe that this is actually supervised learning, as the author is training a classifier on preexisting labels (the genres). I believe that unsupervised learning would not make use of a target variable at all. If the network architectu…

Came here to say this.

Re: Building a Music Recommender with Deep Learning

#47

Post by matchagaucho »

The greatest value of a music recommendation engine, IMO, is cross-genre discovery.

The history of recording industry "Genres" has close ties to cultural segregation. Pandora's Music Genome approach is optimized to break the genre barrier.

It'd be interesting to see how many "Down tempo" songs shared characteristics with "R&B", for example. I think the Author's approach could still be applied.

Re: Building a Music Recommender with Deep Learning

#48

Post by personjerry »

My understanding of convolutions is that it's a way of extracting patterns from images. To convert audio into an image and then create convolutions from that seems... convoluted, if you will. I imagine a better way would be to think of what the equivalent of a convolution would be in the audio space? I.e. noise detection, treble/bass filters, etc.?

Re: Building a Music Recommender with Deep Learning

#49

Post by mordredklb »

This is pretty cool. Maybe I'm missing something, but what's the point in the initial genre training?

He's taking 185000 samples, and finding similar "looking" samples elsewhere in other songs, and then making recommendations based on that. I don't see what that could possibly have to do with genre labels, unless we're under the assumption that finding a match between a Drum & Bass song and one that seems similar with a tag of Trance is somehow a bad match? (which very well could be the case, but seems like a big assumption to make off the bat)

Are these recommendations silo'd to the current genre or are they allowed to span genres?

Re: Building a Music Recommender with Deep Learning

#50

Post by mordredklb »

peteretep wrote: post #11

Tangential: has anyone found anything that doesn't completely suck for recommending books? Goodread's recommendations are terrible.

Agreed. "Oh, you like Horror? Have you heard of Stephen King?"

I've completely given up on goodreads for recommendations and just google "best books 2017" now and usually can find some good lists.

Post reply on HN