Aug 10

What Can You Learn From Last.FM? (Part II)

FT + Proven By Science//5 comments • 811 views

In part 1 of this series looking at artist metrics on Last.FM, I talked about PPL (Plays Per Listener) and also the relative popularity of each act’s top track.

In this part we dig a little bit deeper into an artist’s catalogue, with two more metrics based on their list of top tracks (which, remember, are the tracks with most listeners over the last six months, not over the whole of LFM’s history). I’m calling these metrics – rather unimaginatively – head and body. “Head” is the number of listeners to the tenth most popular track expressed as a percentage of the number of listeners to the first. “Body” is the number of listeners to the fiftieth most popular track expressed as a percentage of the number of listeners to the tenth.

Both of these are based on the same principle – ratios of popular and less popular songs in an artist’s catalogue – but they turn out to measure quite different things. Head measures the extent to which an act is a several-hit wonder. A high head means that your top track isn’t that much more popular than your tenth, which usually means you’ve racked up either a bunch of successful singles or have at least one album that people are keen to listen to in toto. A low head means that you have a few, or maybe just one track which people are particularly keen on but that interest doesn’t extend very far – it suggests a big chunk of casual listeners in your audience.

A high body, meanwhile, means that people are keen to dig deeper into your back catalogue (50 songs is 4+ albums worth for most bands) and a low body either means they’re not, or that you haven’t GOT that kind of back catalogue. It’s usually pretty obvious if this is the case, as the body will be microscopic, since by the time Sleigh Bells or Wavves hit track 50 we’re in the realm of misfiled tags and typos. It certainly isn’t the case that artists who have that kind of back catalogue automatically get a fat body score though.

Head scores of over 50% seem to be pretty good, below 30% suggest the presence of at least one catalogue-outshining hit. In the top bracket you find “album acts” old and new – Pink Floyd are up there, for instance, but so are Vampire Weekend with an enormous 77% head, i.e. their tenth-most-popular track gets more than three-quarters the plays of their most popular one. In other words when someone puts on a VWE record they probably stick it out.

Just below the album acts you find a smattering of pre-Beatles icons – Sinatra, Elvis, Duke Ellington – men with broad enough catalogues to withstand even standout songs: Sinatra’s 10th most popular track gets over half the listeners of “My Way”. Modern pop icons – even those like Madonna with a basket of hits – dip under 50% head scores: there’s always a few songs that are much more popular than their others.

Down at the bottom you’ve got the occasional newbie like Ke$ha (14%) or Drake (9%), but you’ve also got a lot of 80s acts remembered for one or two songs – Dexys, Soft Cell, Bananarama.

Here’s where the relationship between head and body gets interesting though – or rather, the lack of it. The thing about these scores is that they don’t really correlate that much – you can quite credibly have a fat head and a thin body or vice versa. The averages are distorted by acts with small discographies but broadly speaking above 35% seems like quite a big body and below 20% is a thin one (if you’ve got a back catalogue that would merit more, that is.)

So this makes them diagnostically a bit richer, i.e. you can draw up a QUADRANT.

(Positionings in this quadrant are relative rather than reflecting absolute numbers, i.e. I hand-drew it, it’s not an actual chart.)

So let’s have a look at this. In the top-right you have acts with a big head and a big body – lots of popular tracks AND a deep back catalogue. Oh look, it’s the Beatles! But also Yo La Tengo, Radiohead, Kraftwerk, NIN: acts who have devoted fans who see their work as a body rather than a catalogue to be picked from.

Some of the interest lies in who doesn’t make it in: in the top left quadrant you get older acts like Elvis, Queen, and the Doors – people with extensive catalogues but who don’t have the big “body” scores which might be expected. This is the “good for one album” quadrant – whether that’s a debut album or a greatest hits isn’t reflected in the raw stats.

Below that is the “good for one song” quadrant at bottom-left – this section includes a lot of the acts I particularly like, perhaps because I believe people who’ve made one great pop song will often have made more and am happy to delve in on that basis. But the Last.FM public don’t agree with me so here are the Human League, Britney (though she does much better than most pop acts), ABBA on the borderline – lots of hits but no body to speak of – and, interestingly, the Beach Boys.

And on the bottom right is the most intriguing bunch of all – the rarest quadrant, made up of bands with one or a few big hit songs and a bunch of devoted followers who really get stuck in. If you’re happy to listen as far as your tenth Kate Bush or Johnny Cash track, you’re likely to stick around for a lot more. Most extraordinary is Lil Wayne: very very few of the many “Lollipop” listeners dig into Weezy’s labyrinthine back catalogue but a lot of those who do really get stuck in.

In part III I’ll link the data set as a download, point out important caveats, note some quirky results and also draw some conclusions from all this.


  1. 1
    Andrew Hickey on 2 Aug 2010 #

    I was curious about the Beach Boys one (being an obsessive Beach Boys fan) and looking at their results I think I can see what’s happened – they’ve got (almost uniquely I think) a *two album with overlap* popularity – people are either listening to their ‘one classic album’ – Pet Sounds, or to their Greatest Hits, or both. But there are three songs that are on pretty much every BBs hits compilation *and* on Pet Sounds – Wouldn’t It Be Nice, God Only Knows and Sloop John B – and those are massively distorting the results.

  2. 2
    swanstep on 3 Aug 2010 #

    What’s the advantage of relativized head definitions (i.e., #50 listens as % of #10 listens rather than #50 as a % of #1 listens)? That would appear to add a layer of artifacts to the analysis. (% of #1 listens does too but that’s inevitable to get artists on a common scale/comparable so is worth it.)

    More generally, why bother with the head/body division when I assume the same data will support comparison of whole #1-#50 listen tally curves (both before and after common scaling)?

    Oh, and this is as good a time as any for an Onion classic.

  3. 3
    Andrew Hickey on 3 Aug 2010 #

    The head (10/1) allows us to see the relative popularity of (roughly) the artist’s most popular ‘album’ compared to their biggest hit. Body (50/10) allows us to see (very roughly) the popularity of the artist’s fourth- or fifth-most-popular album against their most popular. I imagine that with the exception of the Beatles pretty much every artist would have a nugatory 50/1 value (and even with the Beatles that value is less than 50%), so you wouldn’t get much interesting data out of it, but by doing 50/10 you can see how many fans the artist has of their whole career.

    It allows you to see, for example, artists who have one huge hit but a devoted small fanbase. Take Loudon Wainwright (who I chose as an example before looking at his page and seeing his one hit isn’t actually his one hugely popular song, but he does have one, so it still works) as an example:

    His most popular song, Daughter, has 4623 listens. The tenth most popular has only 866. That tells us that less than one in five people who like his ‘hit’ like his ‘greatest hits’. But the 50th track, Feel So Good, has 294 listeners. Dividing that by the top track doesn’t give us much new information – not many people who like the top track listen to this one, but we knew that. But it *does* say that 34% of those who listen to his tenth most popular song listen to that one, which in turn suggests that he has a few hundred people on last.fm who like him enough to listen to quite a few of his albums.

  4. 4
    Tom on 3 Aug 2010 #

    Yeah I ran 50:1 too but it wasn’t as interesting in terms of segmenting artists (it correlates closely with 10:1).

    If I wasn’t doing this by hand I’d look at curves! Even then I’d probably want to do something like this (split between top end and bottom, but with a better handle on the gradients and when the drop-offs happen). Though now I think about it another way to look at it would be to see how many tracks it takes to get to 25% or 10% of your top score. I think this might just end up telling us about one-hit-wonders though.

  5. 5
    swanstep on 3 Aug 2010 #

    @Andrew, Tom. Thanks for the explanations. Since fast drop-off of the percentages involved seems to be the real problems here, an alternative to relativization might be to use a (negative) log-scale on the abs percentages/fractions. At any rate, if I have time, I may graph a few things from Last in that way and see what visuals emerge.

Add your comment

(Register to guarantee your comments don't get marked as spam.)


Required (Your email address will not be published)

Top of page