Share
Viral statistics and how to take over last.fm's journal system
15 May 2007, 09:23
During the last week, one of my scripts (for computing your "eclectic score") has been spreading to thousands of Last.fm journals, and the posts have taken over recommended articles for reading, artist related journals, ... By just looking at the dynamics of how the posts have spread (I have now generated over 40K user profiles), it feels like I'm handling a deadly virus, although this wasn't my intention at all.
The script contained a couple of properties that allowed it to spread really quickly and take over control of last.fm's journal system:
* The script generates BBCode, which can conveniently be copy/pasted into all kinds of last.fm journals, thread posts, comments, private messages, ... People are lazy, and want something easy. If all you need to do, is to copy the whole journal post in and paste it in a text box at last.fm, people generally seem very willing to write journal posts (probably more than they're used to)
* When the BBCode is interpreted in last.fm, a URL links back to the script, so that people can easily generate their own eclectic score. If people would have to search for the script, they're probably not going to look at all.
* The resulting journal posts (previously) contained a lot (close to 100) artist connections. These artist connections draw a lot of readers to the posts, as the post becomes highly recommended in many circumstances. In the recommended reading, my profile has been spammed with eclectic-score-posts, since they relate to artists that I am interested in reading about. Also, on many artist pages, the related journal entries are currently dominated by eclectic-score-posts, spoiling the fun for journal readers that like to follow-up on their favorite artists. (I have received many rightful comments with respect to this, and it is the reason why I have changed the script's output, see later)
* Lastly, but not at all least, a "score" is computed, which gives the whole "eclectic" review a very competitive edge. There's currently tens of threads on last.fm where people show off with their scores, trying to impress people with a vaguely defined ensemble statistic.
All essential properties of a viral meme seem present here. The eclectic-score-posts have a very high visibility (and thus fitness) and can easily replicate among last.fm's population. The response on my web server's load has also been showing viral dynamics, with an unbounded exponential growth of hits (I think I can assume the growth to be unbounded for now as the script has only reached about 1/1000 Last.fm users).
In order to cut down on the viral dominance of the "eclectic score" I have now removed the related artists from the generated BBCode. Posts that are now being copy/pasted from the eclectic score calculator will no longer make 10s of connections with artists, therefore (hopefully) clearing up the related journals on artist profile pages.
I had lots of fun with the eclectic score scripts, but now I feel that it has completely f*cked up last.fm's journal system. In the end, the eclectic score is quite a strange measure anyway, so I don't feel bad about silencing it for now. Hopefully, other statistics-authors foresee similar problems with future scripts, and react when the scripts get viral on us.
The script contained a couple of properties that allowed it to spread really quickly and take over control of last.fm's journal system:
* The script generates BBCode, which can conveniently be copy/pasted into all kinds of last.fm journals, thread posts, comments, private messages, ... People are lazy, and want something easy. If all you need to do, is to copy the whole journal post in and paste it in a text box at last.fm, people generally seem very willing to write journal posts (probably more than they're used to)
* When the BBCode is interpreted in last.fm, a URL links back to the script, so that people can easily generate their own eclectic score. If people would have to search for the script, they're probably not going to look at all.
* The resulting journal posts (previously) contained a lot (close to 100) artist connections. These artist connections draw a lot of readers to the posts, as the post becomes highly recommended in many circumstances. In the recommended reading, my profile has been spammed with eclectic-score-posts, since they relate to artists that I am interested in reading about. Also, on many artist pages, the related journal entries are currently dominated by eclectic-score-posts, spoiling the fun for journal readers that like to follow-up on their favorite artists. (I have received many rightful comments with respect to this, and it is the reason why I have changed the script's output, see later)
* Lastly, but not at all least, a "score" is computed, which gives the whole "eclectic" review a very competitive edge. There's currently tens of threads on last.fm where people show off with their scores, trying to impress people with a vaguely defined ensemble statistic.
All essential properties of a viral meme seem present here. The eclectic-score-posts have a very high visibility (and thus fitness) and can easily replicate among last.fm's population. The response on my web server's load has also been showing viral dynamics, with an unbounded exponential growth of hits (I think I can assume the growth to be unbounded for now as the script has only reached about 1/1000 Last.fm users).
In order to cut down on the viral dominance of the "eclectic score" I have now removed the related artists from the generated BBCode. Posts that are now being copy/pasted from the eclectic score calculator will no longer make 10s of connections with artists, therefore (hopefully) clearing up the related journals on artist profile pages.
I had lots of fun with the eclectic score scripts, but now I feel that it has completely f*cked up last.fm's journal system. In the end, the eclectic score is quite a strange measure anyway, so I don't feel bad about silencing it for now. Hopefully, other statistics-authors foresee similar problems with future scripts, and react when the scripts get viral on us.
Comments
-
mbatterham wrote:
I think you should change it, so rather than being a numerical score it uses another method to display your eclecity -
Jazz13 wrote:
the script itself seems pretty solid, but i would leave the artists unlinked and leave the option to the author whether they want to add the connections. -
decepti0n wrote:
So it was YOU!
J/k, doesn't bother me and I don't even use it, i've just seen those journals everywhere.
Nice bit of coding though! -
JakobVirgil wrote:
What is the average score?
-Jake -
talking_animal wrote:
Thanks for the hours of fun and excitement I've had perusing eclectic scores for dozens of last.fm users. To build upon mbatterham's idea above, maybe you could generate colors instead of numbers for a non-competitive eclectic score (especially since the color ID's are in hex numbers and hard to compare to each other). -
calebselect wrote:
Don't you think it would be better if instead of just taking the top 5 similars for the top 20 bands, it took the top 4 similars and also added the top 20 to the list?
im just thinking it would be more accurate. cos if someone has limited taste, its unlikely that they would have 20 top bands that are all similar to the same 5 artists (who aren't on the top 20). They're more likely to be similar to the other 19 bands on the top 20.
If you did it this way, the minimum score would be 20.
Anyway good on you for doing this, its interesting. -
D-C-L-X-V-I wrote:
What about linking to artist pages (not with the last.fm (artist)(/artist) tags but something like, (url=http://www.last.fm/music/band/)band(/url)
...I don't care if it's linked or not, just wondering what you might think of my idea.
P.s. I know they used to be linked, and I know and prefer them unlinked.
And of course the parenthesis are in place of the brackets. -
Korosuke wrote:
Great stuff!
Thanks for making the script. =) -
PedroGhandi wrote:
i've used your script and found it lovely (94/100). Since i'm one of those who enjoy the journals i was careful enought to not post the tag cloud in a journal. (i did it on the thread)
since it's now safe, i might just do that.
Is there a way to, if it's posted on a journal, it does not contain links?
Do you plan to apply your cleaning to the other posts?
I liked the idea of D-C-L-X-V-I, specially in the recommended artists tag cloud.
Thx again. -
Green_Dragon wrote:
Hi, man, you are one of that programmers, who add a bit of fun with their scripts, linked to whatever with a data base. Thanks for this ((((((:
That list of SQRTs is interesting, but rater long. There are some graphics with diagramms on your tag-cloud-generating page, so is it possible to make the results of the diagnosis look like that? -
Den_Dolf wrote:
there are ALOT of artists missing!
check my profile :) they are ALL electronic music...
for instance: David West, Signalrunners, sander van doorn, ... -
LynnSnow wrote:
I hadn't even noticed any of this was going on. I'm not one to use the journals on here, I guess. :/ So not everyone is miffed at you. :) Haha. -
freelancelove wrote:
Adorably earnest. Thank you. -
blisterman wrote:
I haven't been on last.fm for ages. Interesting that this has become so popular. I came up with a very similar way of testing eclecticness, last year http://www.last.fm/user/blisterman/journal/2006/05/3/128873/ -
daneeeh wrote:
you should take some advantage of your system and put some kind of advertisement on your site, aren't you interested in doing that? -
Green_Dragon wrote:
viral advertising??? That's efficient, but AWFUL!
-
namnatulco wrote:
Nice solotion. Thanks for considering both parties (users and last.fm's journal system) here. -
PoissonSolluble wrote:
When counting the top 5 similar artists of top 20 artists, it is far too easy to get high result. Mine is 80/100 even though my musical taste is somewhat narrow (I said somewhat).
That's why I'd prefer if it was top 20 similar artists of top 50 artists (possibly divided with 10, minimum being then 2 and maximum 100).
calebselect's idea is good too:
Don't you think it would be better if instead of just taking the top 5 similars for the top 20 bands, it took the top 4 similars and also added the top 20 to the list?
im just thinking it would be more accurate. cos if someone has limited taste, its unlikely that they would have 20 top bands that are all similar to the same 5 artists (who aren't on the top 20). They're more likely to be similar to the other 19 bands on the top 20.
Added to mine result would range between 5 to 100 (wow, the original range :D)
[top 19 similar artists of top 50 artists included with the top 50 artists and divided with 10]
Consider about it! -
aliekens wrote:
Combining above commenters' ideas results in the super-eclectic score!!!
Take your top 50 artists. For each of these artists, collect the top 20 similar artists (where the artist itself is the #1 most similar). The resulting number of unique artists is your super-eclectic score. You can compute your own score at http://anthony.liekens.net/pub/scripts/last.fm/supereclectic.php
My super-eclectic score is currently511/1000
The most similar artists for my profile are Massive Attack (16), The Chemical Brothers (15), Moby (14), Underworld (14), Orbital (14), Leftfield (13), Röyksopp (11), Fatboy Slim (11), Faithless (10), Groove Armada (10) -
Athropos wrote:
I'm not sure I understood how your super-eclectic score works.
Let's say all artists in my top 50 have a different music style, I should get a high score, right? But let's say that all these artists are the only one in their own style (no similar artists), then I'll get only 50/1000? But if there are a lot of artists in each style, let's say more than I can actually listen to, then I'll get a very high score? -
bavanandel wrote:
Okay I see. That's funny, I didn't read the whole eclecticity-post and computed the whole damn thing myself, *including* links to the artist pages. Well, let's just say I'm an infection to the system, haha! -
SNR1984A wrote:
Why don't you just reformulate this script to turn the artist names as being just a static hyperlink?
