Archive for category recommendation
Music Discovery Deathmatch at SXSW
Posted by Paul in Music, recommendation on March 16, 2009
Tomorrow (Tuesday) at 11:30 AM, Anthony Volodkin (creator of the Hype Machine) and I will go head-to-head for 60 minutes in front of a live (and probably hostile crowd) at SXSW for a panel called “Help! My iPod thinks I’m emo” . Anthony and I disagree about many things related to music discovery and recommendation but we do agree on one thing. Current music recommendation is mostly crap. In this session, I’ll be talking about music discovery innovation coming from the researchers, while Anthony will be talking about the new ideas coming from those in the front-lines – the next generation music startups. We have two very different perspectives on the future of music discovery. This will be fun.
Roundtrip tagging
Posted by Paul in Music, recommendation on March 5, 2009
Over the last 5 years, Last.fm has built an incredible database of social tags around music. They have collected millions of short text descriptions of artists, albums and tracks. These tags are a great way to explore for new music, and Last.fm exploits these tags on their site to great effect. But what if you want to use the tags to help you play music from your own collection? Until now you were out of luck – you had to resort to the iTunes style of exploring your personal music collection – resulting in lots of playlists from artists in proper alphabetical order but with no musical cohesiveness. Now, Last.fm has just released a prototype, called Boffin that allows you to use the great body of last.fm social tags to play music in your own collection. The program is called Boffin – I took it for a quick spin and I really like it.
When you run Boffin for the first time, it enrolls your music collection. For me, with about 10K tracks, this took less than 5 minutes. During this time, Boffin is ‘phoning home’ to last.fm to get the tags that have been applied to your artists and tracks. I call this Round Trip Tagging – we give some tags to last.fm when we tag music, and they give lots of tags back to us to let us label our own collection. Once enrolled, Boffin gives you a tag cloud interface to your music collection. Select a few tags, hit the play button and you are listening to your own music. Here’s what my Boffin tag cloud looks like:
Of course, the listening experience is going to be good, because I’m listening to my own music and, presumably, I like that music already.
For a prototype application, Boffin is really well polished (at least the mac version is). While enrolling my music collection, Boffin shows images of all the artists in my collection that it is finding. I was rather amazed at how fast they were able to enroll my collection (I guess Boffin isn’t subject to the rate limits that users of the Last.fm developer API are subjected to). I did find a few times that I thought Boffin had hung up, because I couldn’t select tags anymore, but it turns out that Boffin disables tag selection when it is actually playing music. Once I hit the stop button, I could select tags with no worries. Boffin will even make it easy to generate the popular wordle tag cloud of my personal collection:
Good job to the folks at Last.fm, Boffin is pretty neat!
A recommender comic …
Posted by Paul in fun, recommendation on February 23, 2009
Damn you, Mr. Bezos!
Posted by Paul in fun, Music, recommendation on February 22, 2009
Once again, I was blind-sided by the Amazon recommender. I was placing an order for a few books that my wife wanted. Easy enough, and it would only take 5 minutes. But while I was adding Marie’s books to the shopping cart, a recommendation for a new Keith Emerson CD caught my eye. The last thing I bought by KE was not so good, but the reviews for this CD were rather positive – and so I added it to the cart. And then another Keith Emerson Anthology CD was recommened “just for me” – which has some songs I haven’t listened to for years and are still sitting on vinyl in my attic. That 2 CD set found its way into my shopping cart too.
And then, while at the final checkout, the new Ben Fry / Processing book was sitting there, with 13 excellent reviews. How could I pass that up? And so with an extra $80 removed from my wallet, I finally checked out of the store. Really, that should be illegal. But I’m looking forward to the new tunes and the new book.
In which I am ridiculed for my music tastes …
Posted by Paul in fun, Music, recommendation on February 16, 2009
I was giving a talk last week about music recommendation at a local college. I was explaining how some of the various online music recommenders work when I noticed that some of the students were chuckling and laughing. I had checked my fly right before I started talking so I knew it wasn’t that. Then some wise guy in the front row made it all clear: “Do you really like to listen to Hilary Duff?”. After a moment of confusion, I realized that I was showing my Pandora radio stations that included the second most infamous Hilary.
My Pandora Radio stations
I tried to explain that I sometimes listen to my Pandora radio with my 13 year-old daughter. I’m not sure that they really believed that.
This evening, my daughter and I were having dinner and talking about music. She’s past the Hilary Duff and Hannah Montana phase. She’s moved onto the Veronicas (check out her latest review) – so we listend to a bit of Veronicas’ radio on Last.fm – which has now been faithfully scrobbled as part of my listening history forever:
I do like listening to music with my daughter. She knows all of the artists, and (seemingly more important), all the back stories, interconnections, failures and gossip about the artists. That seemed to be as important as the music itself. And although it is fun to listen to with my daughter, the music is not really to my taste. I do want to make it clear to anyone, whether it is a class at the local college or a potential future employer that I’m not really that into bubblegum pop.
Twisten.fm – Music gone viral.
Posted by Paul in Music, recommendation on February 13, 2009
The fine folks over at Grooveshark have just released Twisten.fm. Like Blip.fm, Twisten.fm combines micro-blogging and music, and like Blip.fm, instead of creating an entirely new microblogging network, Twisten.fm piggybacks on top of the existing Twitter network.
With twisten, you can make tweets with just about any song in the world (Grooveshark has millions of tracks in its catalog) – twisten will automatically generate a twisten tiny url to the song, and if any of your twitter followers click on the link, they are taken directly to the song page on Grooveshark where they can listen to the song.
twisten a song
Here’s what the song tweet looks like on twitter:
Twisten makes it easy for you to tell your twitter followers that you like a song – with Twisten, you just type the name of the song, and twisten does all of the hard work of finding the song in the catalog, generating a tiny URL for the song and posting it to Twitter.
Twisten also makes it easy to listen to music posted by others. If you use the Twisten web app, you can easily listen and browse all of the twisten tweets of your followees or the world at large.
Listening to Twisten tweets
With the twisten app, you just see the Twisten tweets, which makes it a perfect app for browsing through new music. It is easy to listen, since the player is embedded right in the page. You can also listen to the music that is being posted by everyone.
I suspect that Twisten.fm is going to be a really big deal. First and foremost, it is an incredibly viral app. Just by using Twisten, you are telling all the world about it. 18 hours after it’s release, Twisten is #6 on the list of Twitter trending topics.
Second, it doesn’t re-invent the wheel. Instead of building a whole new social network, it sits on top of Twitter, one of the largest existing social networks – it doesn’t have to build up a network from scratch.
Twisten is really neat, I like it a lot – still, there are a few places where it could be improved.
First of all, when listening to music on the Twisten site, the music should never stop when I navigate to a different part of the site. Right now, if I’m listening to a particular tweet, and decide to check out what ‘everybody is listening to’, the music stops. The main Grooveshark app does a much better job of keeping the music playing all the time whilst one navigates through the site.
Currently, when I click on a twisten tiny url in twitter to listen to a song, instead of taking me to Twisten, the URL takes me to Grooveshark. I understand that Grooveshark is hosting all of the music, but it seems to me that if you want to really make Twisten go viral, the links should bring listeners straight to Twisten, where they can listen to the music, and while there, start Twisten their own tweets.
The listening experience on Twisten is a hunt-and-peck style. I see a song, I click on it, I listening to it, and then I go and find the next song. That’s fine when I am exploring for new music, but if I just want to listen to music, I’d like to be able to turn Twisten into a radio station, where I listen to the music that my friends have been twittering. Ideally, I should be able to listen to tweets all day without having to click a mouse button. TheSixtyOne does a great job of keeping the music flowing. Twisten should follow their model.
I wish Twisten.fm would scrobble all my tweets and listens – it’d be great if every music app in the world scrobbled my listening behavior.
Twisten is able to collect all sorts of interesting information about who is listening to what music. I hope they do some interesting things with this data. For instance, they could create a Twitter Music Zeitgest that shows the songs and artists that are rising, popular, or falling. Since Twisten knows what I’ve been listening to, and what I like (because I can ‘favorite’ twisten songs), Twisten should be able to connect me up with other Twisten listeners that have similar tastes so I can use their twitters and listens to guide my own listening. Twisten is going to be able to collect lots and lots of user listener data, so it should be interesting to see what they do with it all.
Twisten has the potential to be the real breakout music application of 2009. It has all the ingredients – a huge catalog of free music, and a viral model that leverages one of the largest and most active social networks. When iLike released it’s facebook app, iLike became the fastest growing music app ever, adding 3 million users in two weeks. Twisten has a good chance to do the same thing.
Music Recommendation and Discovery in the Long Tail
Posted by Paul in Music, music information retrieval, recommendation on February 12, 2009
Over the last couple of years, I’ve been lucky enough to get to know Music Information Retrieval researcher Oscar Celma. Oscar and I collaborated on a tutorial on music information retrieval that we presented at ISMIR 2007. We spent many, many hours on phone, email and IM sifting through every aspect of music recommendation.
This fall, Oscar completed his PhD Thesis. Oscar asked me to be the ‘external reader’ so I spent a good part of my Christmas break reading and re-reading the 230 page thesis. Oscar really has done a phenomenal job at looking at the issues and problems in music recommendation and in particular how they (or more accurately, how they don’t) help you find music in the long tail. Oscar’s analysis of how far different types of recommenders can push you deep into the tail.
Oscar has just published he’s thesis along with some supplementary info and code on the web site: Oscar Celma PhD. If you are involved in Music 2.0, I highly recommend reading it.
Some cool plots:
3D Representation of the long tail
And the abstract …
ABSTRACT
Music consumption is biased towards a few popular artists. For instance, in 2007 only 1% of all digital tracks accounted for 80% of all sales. Similarly, 1,000 albums accounted for 50% of all album sales, and 80% of all albums sold were purchased less than 100 times. There is a need to assist people to filter, discover, personalise and recommend from the huge amount of music content available along the Long Tail.
Current music recommendation algorithms try to accurately predict what people demand to listen to. However, quite often these algorithms tend to recommend popular -or well-known to the user- music, decreasing the effectiveness of the recommendations. These approaches focus on improving the accuracy of the recommendations. That is, try to make accurate predictions about what a user could listen to, or buy next, independently of how useful to the user could be the provided recommendations.
In this Thesis we stress the importance of the user’s perceived quality of the recommendations. We model the Long Tail curve of artist popularity to predict -potentially- interesting and unknown music, hidden in the tail of the popularity curve. Effective recommendation systems should promote novel and relevant material (non-obvious recommendations), taken primarily from the tail of a popularity distribution.
The main contributions of this Thesis are: (i) a novel network-based approach for recommender systems, based on the analysis of the item (or user) similarity graph, and the popularity of the items, (ii) a user-centric evaluation that measures the user’s relevance and novelty of the recommendations, and (iii) two prototype systems that implement the ideas derived from the theoretical work. Our findings have significant implications for recommender systems that assist users to explore the Long Tail, digging for content they might like.
Music Machinery 