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

21–30 of 74 posts

Re: Building a Music Recommender with Deep Learning

#21

Post by arnioxux »

amelius wrote: post #15
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…

What could also help in this case is "collaborative filtering", [1]. [1] https://en.wikipedia.org/wiki/Collaborative_filtering

Which just means "people similar to you who like x also like y".

This does make a lot more sense than analyzing the audio of the music IMO. For example youtube does this okay and if you look for a Mazzy Star song after watching Ricky and Morty (a tv show), it will recommend other Ricky and Morty soundtracks even if the style is completely different. This isn't something you can predict with just audio data.

Re: Building a Music Recommender with Deep Learning

#23

Post by mattmcknight »

oscii wrote: post #18

In my opinion, the results are not quite exciting as they might seem like at the first glance. The hip-hop and minimal house classification perform almost randomly (the random classifier would have accuracy of 50%). The claim of music genre subjectivity is not fully appropriate for the categories used in this work: the presented genres are quite distinct, and they have objective differences. Knowing only BMP and rhyt…

"The hip-hop and minimal house classification perform almost randomly (the random classifier would have accuracy of 50%). " You are assuming that this is a series of binary classifiers. It is multiclass classification, so the base rate for nine classes is 11%.

Re: Building a Music Recommender with Deep Learning

#24

Post by oscii »

mattmcknight wrote: post #23
oscii wrote: post #18

In my opinion, the results are not quite exciting as they might seem like at the first glance. The hip-hop and minimal house classification perform almost randomly (the random classifier would have accuracy of 50%). The claim of music genre subjectivity is not fully appropriate for the categories used in this work: the presented genres are quite distinct, and they have objective differences. Knowing only BMP and rhyt…

"The hip-hop and minimal house classification perform almost randomly (the random classifier would have accuracy of 50%). " You are assuming that this is a series of binary classifiers. It is multiclass classification, so the base rate for nine classes is 11%.

Oh yes, you're right. Thanks for the correction! A recent peer review is still in my head.

Re: Building a Music Recommender with Deep Learning

#25

Post by visarga »

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.

This is a content based approach. Music suggestions are content based, user based or combined.

Re: Building a Music Recommender with Deep Learning

#26

Post by flashman »

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…

> Popular songs (i.e. if almost everyone has a Beetles album in their playlist, getting "people who bought this also bought" recommendations for anything would list Beetles

I've been learning recommendation engines by looking at peoples' Steam games libraries.

One feature of the data set is that many, many people own multiple versions of Counter-Strike as well as Team Fortress 2. So "a high number people who bought [almost any game] also bought Counter-Strike: Global Operations" is a recurring problem with a naive recommender.

What I've been learning how to do is weight recommendations by how 'surprising' they are, for want of a more accurate term. If 80% of people who own Game A also own Game B, but only 5% of the total population owns Game B, then we should upweight that relationship.

Re: Building a Music Recommender with Deep Learning

#27

Post by Pigo »

madmax108 wrote: post #5

This is interesting, but fairly easy to confuse. Esp. would be interesting to see what results come up when you use modified "artistic" spectographs like that of Windowlicker by Aphex Twin [1]. One thing I've learned from years of having worked with audio and images is that image representations of audio are horrible representations of it (other than for temporal changes). The results are good though! Good work! :D […

It also doesn't help that music from every genre is becoming more homogenized as time goes by [1]. If your comparing by similarity, then this is only going to get more difficult.

[1] http://journals.plos.org/plosone/article?id=10.1371/journal....

Re: Building a Music Recommender with Deep Learning

#28

Post by mattmcknight »

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…

Right, in a unsupervised model (say k-means), it divides the data into 9 similar groups and then it is up to you to label what those 9 groups are.

Re: Building a Music Recommender with Deep Learning

#29

Post by visarga »

emadb wrote: post #13
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…

I totally agree with you. Sometimes you love a song from the first listen, it bites you even if came from an artist that you don't know. My dream is a suggestion engine that "examine" the melody, the harmony, the frequencies that make the song and finds songs that are similar based on that parameters. Probably a signal analysis could help in finding why you like that songs.

But this article does exactly that: signal analysis with neural nets on song spectrograms. It just doesn't generate the kind of matches you want.

Re: Building a Music Recommender with Deep Learning

#30

Post by bsenftner »

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 all in the same mix and it works because the lyric content is different takes on the same things. In fact, new friends are sometimes dizzy from my music choices, and then at some point they hear the thematic concept of my mixes and get it.
Post reply on HN