The Last.FM Similarity Project

Anteater · 10-11-2009, 12:32 PM

+81 - +8 - 1
adidass - AD - 2
anticipation - AP - 3
Bardonodude - BA - 1
Bulldog - BD - 1
Captain Caveman - CC - 1
dreadnaught - DN - 2
jackhammer - JH - 3
Kayleigh - KL - 1
khfreek - KF - 6
LoathsomePete - LP - 2
Molecules - MC - 2
NumberNineDream - ND - 2
sidewinder - SW - 4
storymilo - SM - 2
toretorden - TT - 6
Urban Hatemonger - UH - 1
wolverinewolfweiselpigeon - WW - 2
zeppy111 - Z1 - 5

Man, bitches don't know 'bout my prog.

khfreek · 10-11-2009, 12:38 PM

I became a 6 with Anteater in the past few days... is that gonna be bad for the data?

Guybrush · 10-11-2009, 01:01 PM

Quote:

Originally Posted by khfreek

I became a 6 with Anteater in the past few days... is that gonna be bad for the data?

No worries, I can fix it

simplephysics · 10-14-2009, 06:30 PM

Quote:

Originally Posted by toretorden

Come on Anteater and Bardonodude .. Gimme your data!

Bump? Any chance we can get the ball rolling with this, I'm rather curious as to how it will turn out.

Guybrush · 10-14-2009, 06:44 PM

I left a message on Bardono's profile, but no data yet.

Guybrush · 10-15-2009, 08:19 AM

Okay Bardonodude, you had your chance. I've now done the analysis without your data

Here is the matrix :

I've played around with your numbers a bit. The values and scores now reflect eachother like this :

Very Low = 0
Low = 20
Medium = 40
High = 60
Very High = 80
Super = 100

The matrix should be easy to read. If you want to know how similar Jackhammer and Urban Hatemonger are in tastes, just look to Urban's row, then follow it towards the right until you get to Jackhammer's column. The value there is 80 - which is very high similarity.

The Results :

I used the program "Primer" to do the analysis. It was one of hundreds that could do the job, it seems, but I had to use the demo version which wouldn't let me copy & paste. I had to enter everything manually so there could be a typing error, but I think not. Anyways, on with the show :

Above, you can see a clutter analysis dendrogram. Along the Y axis you have similarity and the lower you have to go through the branching to get from one group to another, the less similar they are. As you can see, Kayleigh forms her own group and is quite the outlier. Basically, her tastes are dissimilar to everyone elses.

I can explain a score here in case someone's confused. Let's look at sidewinder (SW). His most similar neighbours are me (TT) and Zeppy111 (Z1). However, he's closer related to me than he is to Zeppy111. The reason is because Zeppy is also closely related to LoathsomePete who sidewinder has a comparatively poor score, 40, with. The average between the two becomes 60. sidewinder has a better averaged score with me and my closest neighbours, Anteater (AE) and Khfreek (KF) (similarity score of ~67), so he branches closer to our group (TT, AE, KF).

I'll explain another. The group made up by MD and AP have a similarity of 30 to the group made up by CC and +8 - that's how low you have to climb down on the branches to get from one group to the other. The reason it's 30 is members of these groups have a score of 40 with one of the other group and 20 with the other. The average score (20+40/2) then becomes 30.

I have a similarity score of 90 to the group made up by Anteater and Khfreek - because my score is 100 with Anteater and 80 with Khfreek - with the average between them being 90.

I also did some ordination and here's the MDS analysis :

This one is perhaps a little harder to read and (imo) not as good, but the idea is similar as for the dendrogram. The closer you are to someone else, the more similar your tastes are. For example, my tastes are more similar to Anteater's (100) than Kayleigh (0) and Captain Caveman (0). Note that CC is still closer to Anteater than KL is and this is because CC is more similar to people who Anteater also have similarities with. KL is really the lone wolf here.

The problem with this MDS plot is that it has to present the results visually and we don't relate well to eachother on a 2D chart like this so there's going to be errors. How close should you be to someone you are super similar to but who's closest neighbour you are very low with? Basically you can't make it perfect and the stress value indicates how much error, but instead of worrying about the stress value let's just say it's there and so I prefer the dendrogram.

That's it I guess .. any comments? I'm not a statistics master so feel free to tear me a new one if you think I did a bad job. I wanna learn how to do this kind of stuff which is part of why I thought of doing this.

Who knows? Maybe the results were inspiring to some?

By the way, if you have a hard time understanding them, just ask!

Guybrush · 10-15-2009, 10:28 AM

No comment on the results yet?

Looks like me, Zeppy, LoathsomePete, Sidewinder, Anteater and Khfreek should have a party together. As a group, we're reasonably compatible - at least when it comes to taste in music.

Kayleigh, you're on your own!

Molecules · 10-15-2009, 10:36 AM

really swell (today's anachronism) job Tore, it might be cool to someday have all the regulars participate. Although last.fm compatibility scores are prone to change quite often, sometimes with no apparent reason...
I particularly dig the group average chart, very clever

jackhammer and urban hatemonger = dr. jekyll and mr. hyde?

dollarsandcents · 10-15-2009, 10:43 AM

I'd have liked to have been involved with this, shame! Great analysis and statistics though, interesting. Although the tree reminds me far too much of my university dissertation thesis, which involved a lot of fiddling round with DNA and plotting phylogenetic trees of maternal inheritance in Southeast Asia and the Pacific. Urgh, those programmes were fiddly and inconvenient!

Guybrush · 10-15-2009, 10:43 AM

Quote:

Originally Posted by Molecules

really swell (today's anachronism) job Tore, it might be cool to someday have all the regulars participate. Although last.fm compatibility scores are prone to change quite often, sometimes with no apparent reason...
I particularly dig the group average chart, very clever

jackhammer and urban hatemonger = dr. jekyll and mr. hyde?

Hehe, I have a reasonably good score with you as well, but your statistical partnership with Mr. Anticipation ruins things a bit

Aside from Kayleigh, you two seem to be the biggest deviants.

10-11-2009, 12:38 PM	#82 (permalink)
khfreek What a guy Join Date: Oct 2008 Location: Brentwood, TN Posts: 2,123	I became a 6 with Anteater in the past few days... is that gonna be bad for the data? __________________ last.fm

10-14-2009, 06:44 PM	#85 (permalink)
Guybrush Juicious Maximus III Join Date: Nov 2008 Location: Scabb Island Posts: 6,525	I left a message on Bardono's profile, but no data yet. __________________ Something Completely Different

10-15-2009, 08:19 AM	#86 (permalink)
Guybrush Juicious Maximus III Join Date: Nov 2008 Location: Scabb Island Posts: 6,525	Okay Bardonodude, you had your chance. I've now done the analysis without your data Here is the matrix : I've played around with your numbers a bit. The values and scores now reflect eachother like this : Very Low = 0 Low = 20 Medium = 40 High = 60 Very High = 80 Super = 100 The matrix should be easy to read. If you want to know how similar Jackhammer and Urban Hatemonger are in tastes, just look to Urban's row, then follow it towards the right until you get to Jackhammer's column. The value there is 80 - which is very high similarity. The Results : I used the program "Primer" to do the analysis. It was one of hundreds that could do the job, it seems, but I had to use the demo version which wouldn't let me copy & paste. I had to enter everything manually so there could be a typing error, but I think not. Anyways, on with the show : Above, you can see a clutter analysis dendrogram. Along the Y axis you have similarity and the lower you have to go through the branching to get from one group to another, the less similar they are. As you can see, Kayleigh forms her own group and is quite the outlier. Basically, her tastes are dissimilar to everyone elses. I can explain a score here in case someone's confused. Let's look at sidewinder (SW). His most similar neighbours are me (TT) and Zeppy111 (Z1). However, he's closer related to me than he is to Zeppy111. The reason is because Zeppy is also closely related to LoathsomePete who sidewinder has a comparatively poor score, 40, with. The average between the two becomes 60. sidewinder has a better averaged score with me and my closest neighbours, Anteater (AE) and Khfreek (KF) (similarity score of ~67), so he branches closer to our group (TT, AE, KF). I'll explain another. The group made up by MD and AP have a similarity of 30 to the group made up by CC and +8 - that's how low you have to climb down on the branches to get from one group to the other. The reason it's 30 is members of these groups have a score of 40 with one of the other group and 20 with the other. The average score (20+40/2) then becomes 30. I have a similarity score of 90 to the group made up by Anteater and Khfreek - because my score is 100 with Anteater and 80 with Khfreek - with the average between them being 90. I also did some ordination and here's the MDS analysis : This one is perhaps a little harder to read and (imo) not as good, but the idea is similar as for the dendrogram. The closer you are to someone else, the more similar your tastes are. For example, my tastes are more similar to Anteater's (100) than Kayleigh (0) and Captain Caveman (0). Note that CC is still closer to Anteater than KL is and this is because CC is more similar to people who Anteater also have similarities with. KL is really the lone wolf here. The problem with this MDS plot is that it has to present the results visually and we don't relate well to eachother on a 2D chart like this so there's going to be errors. How close should you be to someone you are super similar to but who's closest neighbour you are very low with? Basically you can't make it perfect and the stress value indicates how much error, but instead of worrying about the stress value let's just say it's there and so I prefer the dendrogram. That's it I guess .. any comments? I'm not a statistics master so feel free to tear me a new one if you think I did a bad job. I wanna learn how to do this kind of stuff which is part of why I thought of doing this. Who knows? Maybe the results were inspiring to some? By the way, if you have a hard time understanding them, just ask! __________________ Something Completely Different

10-15-2009, 10:28 AM	#87 (permalink)
Guybrush Juicious Maximus III Join Date: Nov 2008 Location: Scabb Island Posts: 6,525	No comment on the results yet? Looks like me, Zeppy, LoathsomePete, Sidewinder, Anteater and Khfreek should have a party together. As a group, we're reasonably compatible - at least when it comes to taste in music. Kayleigh, you're on your own! __________________ Something Completely Different

10-15-2009, 10:36 AM	#88 (permalink)
Molecules daddy don't Join Date: Feb 2008 Location: the Wastes Posts: 2,577	really swell (today's anachronism) job Tore, it might be cool to someday have all the regulars participate. Although last.fm compatibility scores are prone to change quite often, sometimes with no apparent reason... I particularly dig the group average chart, very clever jackhammer and urban hatemonger = dr. jekyll and mr. hyde? __________________ [SIZE="1"]Eff em tumble her

10-15-2009, 10:43 AM	#89 (permalink)
dollarsandcents more tea vicar? Join Date: Aug 2009 Location: England Posts: 193	I'd have liked to have been involved with this, shame! Great analysis and statistics though, interesting. Although the tree reminds me far too much of my university dissertation thesis, which involved a lot of fiddling round with DNA and plotting phylogenetic trees of maternal inheritance in Southeast Asia and the Pacific. Urgh, those programmes were fiddly and inconvenient! __________________ My Blog: You And Whose Army?