Just over a year ago the Music Information Retrieval team here at Last.fm embarked on a project to see how well we might be able to identify musical characteristics of songs by a process of automatic analysis. Our aim was to fill in some of the gaps left by our existing tagging system.
Last.fm tags make up an astonishing encyclopedia of descriptions, and are a testament to the generosity, knowledge and enthusiasm of our community of users. Together with scrobbles, tags help us power recommendations, radio, and many of the most interesting services that we offer. Although you can make up any tag you like, we noticed that in practice most people use tags that describe genre, or closely related things such as the era or nationality of an artist. On the other hand tags rarely describe the sound of songs in musical terms, and they talk about subjective things like mood less often than you might imagine, given the close connection that most of us experience between music and our feelings about life.
The potential benefits of having a new and separate strand of information about music were obvious, but the big challenge for this project was that existing methods of automatic music tagging simply didn’t work very well. Nine blog posts, two published research papers, three public and numerous internal demos, several hack days, and a great many man hours later, we think we’re starting to get somewhere, and we’d like to show you some results.
As a first taster we’ve put together a visualization of your musical mood over the past 120 days, based on automatically computed machine tags for the tracks which you’ve scrobbled during that time. While individual tags are still far from perfectly accurate, we think that when taken together over all your listening week by week they still paint an interesting picture – one that stands a chance of reflecting real changes in your musical life. Enjoy, and please let us know if you find them interesting!
Comments
Jesse
30 August, 13:54
This would be better. Much better (bottom of page):
http://www.leebyron.com/what/lastfm/
Graham
30 August, 14:18
Jesse,
OK. There you go:
http://playground.last.fm/demo/listeningtrends
Daryl
30 August, 18:40
This would be cool. How can I find out what’s “Happy” vs “Sad”?
Graham
31 August, 09:34
Daryl,
At the bottom of the page on the Mood Report is a link to see samle tracks.
Or, just replace your username in the link below.
http://playground.last.fm/demo/moody/samples?user=USERNAME&days=120
Julian
31 August, 10:58
Regarding the playground page for this:
http://uxdesign.smashingmagazine.com/2012/06/20/links-should-never-say-click-here/
Graham
31 August, 11:27
Julian,
I’m sure we’ll amend that.
It is just our “playground”, so yes, hands-up, we may not employ best practice ;)
Thanks for pointing that out.
dex
31 August, 16:26
Interesting addition to the playground tools for sure – although I wish you would focus to the layout issues than putting out tools only handful of people will ever use (or are even aware of). . .
Anyway, as it sounds like this will take into account user tags of the tracks/albums/artists, can you tell which kind of tags are included so that the users could improve the results – like does it only use simple mood/descriptive words such as “happy”, “sad”, “fast” etc. or also more complex ones, such as long phrases?
Graham
31 August, 16:55
Hi Dex,
It’s not social tagging by users, it’s an experiment in audio analysis, we’re teaching the machine to classify music so that we can improve recommendations.
Now, where did I put that layout file? Ah, there it is… Dex.psd
adam
3 September, 18:03
wow, apparently because I listen to metal, my results come back as nonstop “high energy”, and sad is like a little tiny blip at the bottom of the page. that’s like, straight up worse than any other auto-tagging i’ve tried.
Mark Levy
4 September, 11:17
Hi adam,
That sounds interesting. What would you expect to see given your pattern of listening? Are you able to say which other auto-tagging systems you’ve used and how they were better?
Also, if you didn’t do it already, please feel free to visit http://playground.last.fm/demo/moody/samples and give us feedback about tracks that are badly labelled. That will (gradually) help us make improvements to the system. Thanks!
snyde1
4 September, 18:45
Um, I think I broke it …
Okay, I likely didn’t, but it ain’t workin’ for me.
adam
4 September, 21:13
@mark – metal has moods just like any other genre – I did go through the samples and marks some of them. I was just bummed that some major trends (doom metal, depressive/atmospheric black metal) net the same descriptors (high energy, fast) as stuff that’s actually upbeat (grindcore, thrash & folk metal).
Mark Levy
6 September, 14:50
@snyde1: I can see your report fine at http://playground.last.fm/demo/moody?user=snyde1, might be worth just trying again.
@adam: thanks for the feedback and the details you just mentioned, we’ll try to pay attention in our next round of testing.
snyde1
6 September, 19:07
Mark;
It’s not worked for me yet. The image does not load, and the detail pages (E.g.
http://playground.last.fm/demo/moody/plot?user=snyde1&days=120&pdf=yes) return:
Not available
Whoops, this page is not available right now. Please try again later.
Will try again.
Mark Levy
6 September, 20:53
@synde1 apologies, it was down briefly this evening, it should be working again now – please do try again!
snyde1
6 September, 21:50
Mark;
IT WORKS! Thanks.
I am apparently happy and high energy. People will be surprised.
Snyde.
Anthony DeRobertis
7 September, 15:15
Under “sad”: Peter Ilyich Tchaikovsky: Marche Slave, Op. 31
hmmm, yeah, going to need some work. But interesting!
It’d be nice if I could suggest correct categorizations, not just agree/disagree. And also, if there were to update my report.
Jurij
11 September, 12:15
As for me – Lacrimosa – Satura is the best song for this day. It displays my current mood better than all.
Dan
13 September, 11:09
Just listen to some quality music, it helps :)
Mohit
17 September, 16:48
it works for me
Nethemas
19 September, 00:49
These things are all interesting but when the bleep are you going to properly separate artists? Any kind of “station” based upon artist names is completely and utterly hosed. Listen to JPop artists, but randomly some Latin American stuff floats in because they happen to have the same name. Expect the Japanese synth voice artist Supercell, get an American grunge band. It jarring, it disrupts your listening, it’s annoying.
juancvas
28 September, 15:09
Works for me too.Nice
graveris
28 September, 22:11
That sounds interesting. What would you expect to see given your pattern of listening? Are you able to say which other auto-tagging systems you’ve used and how they were better?
Comments are closed for this entry.