New taste-o-meter new neighborhood


8 Nov 2006, 19:08

Hi everybody!

The new taste-o-meter is online. And the new neighborhood too!
Now the two things are directly related, so it should not happen that you get "medium" or "low" on your neighbor anymore.

How it works? Well, the whole neighbor computation is derived from a rather standard user-based collaborative filtering technique, with some additional "tricks". Bwahahahahah! ;)

This technique returns a (real) value between 0 and 1, that represent the "compatibility" between two users. Since it is rather uncommon that two users share exactly the same set of artists, that number is generally between 0.4 and 0, with a lot of variations for each user.

If we order these values for each user, we get the neighbors. To make an example, here is the list of my neighbors with the computed value:

The value is used as the basis for the taste-o-meter.
Since the top value varies a lot depending on how obscure your tastes are, we cannot use an absolute value to calibrate the taste-o-meter. Rather I coded a simple function that takes into consideration your first neighbor as base, and scale the other values accordingly.

The drawback of this technique is that the resulting match is not symmetric, that is we could have the case in which:
A->B : super
B->A : medium

This follows the principle of: "he is my best friend, but for him, I am only a good friend".

From a logical point of view this sounds correct, but some good hearted people here at believes that we should make all friendships strong, and go for a symmetric version that return max(A->B, B->A) for both people and make everybody (?) happy. Tell me what you think... :)

One last - but very important - note:
Both neighborhood and the taste-o-meter are based on a 6 months dataset. If you (or your neighbor) did not scrobble nor listen to radio during that period.. well, no cookie! ;)

We will switch to 12 months if we see this is not working.

For now it's all, and sorry for my Swiss English! ;)


  • Fick6969

    Awesome job on the revisions! I like the changes a lot. Thanks for all your hard work.

    8 Nov 2006, 19:38
  • shelanamiya

    thank you!

    8 Nov 2006, 19:46
  • Krakowiaczek

    To be honest, i don't like new neighbourhood system. My musical taste is rather wide, ranging from Metal, to Polish Sung Poetry and Soundtracks. I also have quite a lot tracks scrobbled, 21k to be accurate. Today I checked my neighboorhood, and poof... there are no more people who listen to such range of artists, but people who has scrobbled like 1k tracks at all, and most of them were scrobbled by me (they listen to rather narrow part of what I do listen). It's true, that I should be their neighbour, when i listened to ALL their tracks at least once, but why I have THEM as my top neighbours? I Think Tags and Similar artists should be used when calculating neighbours. In my case, i think my neighbours should be people whose musical taste is 30% artists similar to Jeremy Soule, 33% artists similar to Jacek Kaczmarski, and 37% artists similar to Iron maiden, not just similar to one of my top artists.

    8 Nov 2006, 20:04
  • Z_e_t_a

    Switch to 12 months, if not more time, please... (I apologize if this is not the place to write this)

    8 Nov 2006, 20:09
  • nova77LF

    @ Krakowiaczek, you're damn right. Thank you for pointing that out. It is rather strange, because generally users with the same amount of scrobbling get paired, but this is definitely not your case. :P I will work on it tomorrow and see if with some tweaking we can make something. @ Switch: yeah, we will likely switch to 12 and make a separate section for recent neighborhoods.

    8 Nov 2006, 20:20
  • nova77LF

    @creepy_bella: this might be because the dataset I used is about three weeks old. I am generating new data right now (based on 12 months data), and should be ready tomorrow.

    8 Nov 2006, 21:03
  • Krakowiaczek

    Thank you for your help nova77LF.

    8 Nov 2006, 21:09
  • Jazz13

    Why am I getting so many 'Unknown' readings?

    8 Nov 2006, 21:34
  • dontquoteme

    so you're Swiss! all through that entry i was trying to work out where you were really from - thanks for clarifying :). taste-o-meters aren't working at the moment i don't think... everyone comes up as 'unknown'.

    8 Nov 2006, 21:36
  • RJ

    sorry, i had to revert the tastometer back to the old algorithm for now, because the new one was causing some weird problems :( I'll sort it out in the morning. New neighbours are still up tho, and much better in my opinion :)

    8 Nov 2006, 22:04
  • Podex

    I agree, the new neighbours are much better. The text in the taste-o-meter now always says 'You share a few artists in common'. Wasn't it 'a lot of artists' when the compatibility was high or super?

    8 Nov 2006, 22:37
  • nova77LF

    Indeed there are goods and bads about the new neighborhoods. That's just the first step, now its time for some tweaking.

    8 Nov 2006, 22:52
  • nova77LF

    From a quick analysis, it seems that most of the complaints are due to the hidden long tail. The new algorithm takes much more into consideration the artists that are beyond the top 100. This might result, in some cases, in a compromise that looks unfamiliar to the top chart.

    8 Nov 2006, 23:13
  • rpfile

    also in my case i preferred my old neighbors. i found a lot of music i liked by looking at their top 5-10 artists; usually the top 2 artists were common. now my new neigbors really don't seem that similar on the top artists and subsequently i don't trust their musical taste as much. thanks for listening :)

    9 Nov 2006, 1:40
  • nova77LF

    @nyu: The reason is that we had some issue with the new taste-o-meter and I had to be taken down. As soon as it will be up again with the new data these discrepancies will disappear. @Sanashi: There are other reasons for using all the list (related to the generation of the radio playlist), but it is true that for the neighbors this might result in a worst (at a first glance) match. Luckily switching to a top-only dataset is rather easy. Expect new data very soon.

    9 Nov 2006, 7:59
  • transcendentalA

    I'm not too happy with the new predictions of the taste-o-meter. E.g. I get a very high with trident, where I know about 1/3 of his top 500 list and a low with Diotime whereas I see a much higher compatibility for songs played more then 10 times. Is this due to the different amount of time registered and so more tracks scrobbled? Are artist in the top 500 played less then x taken into consideration? In my friends list I miss my former top neighbour Boozi whom's taste I consider allmost identical to mine (that's really rare).

    9 Nov 2006, 8:09
  • bcaptured

    well, i'm the top neighbour of my top neighbour now, but our taste-o-meter compatibility is medium. hope this will change after you get it back again. sorry to hear that you had to take it down again and good luck with it.

    9 Nov 2006, 8:28
  • Extreme_One

    I find it odd that I'm matched as Super compatible with several users on the site yet nobody has me as a neighbour. Oh well, life is odd sometimes ;)

    9 Nov 2006, 10:16
  • Extreme_One

    Please forget the above comment ^ I've just noticed I have [b]new neighbours[/b]. And I'm now in some of their lists as a neighbour too. :D

    9 Nov 2006, 12:27
  • Greg_x7

    Among my top 12 neighbours many people hadn't been on the site for months and/or their last scrobbled track was from about 2 or more months ago, plus they had listened to 0,5k-5k tracks. My point is, should users who are not active be considered neighbours? PS. The compatibility for me is quite good, my neighbours listen to similar music as me and we often share artists in top 50.

    9 Nov 2006, 13:51
  • Deebster

    My new batch of neighbours seem to be pretty good, although there's some unexpected results (like one of my friends who frequently listens to my radio is #2 on my list whereas I'm only 32nd on hers). Of course neighbours shouldn't be symmetric - don't fudge the stats because some people are confused by the results. It's political correctness gone mad! etc... ;)

    9 Nov 2006, 14:49
  • nova77LF

    Ok, the new taste-o-meter is back online. For an updated version of the neighborhood you'll probably have to wait until tomorrow.

    9 Nov 2006, 14:58
  • nova77LF

    @creepy_bella: your problem is related to the data, not to the taste-o-meter. Please be patient, I am working 110% on it.

    9 Nov 2006, 15:19
  • mieteck

    I don't get it. How is it possible: yesterday I was No.1 neighbour of one of my friends. He was my 7th neighbour. We share over 300 artists in common (prooved by beta ver. of taste-ometer which used to show the number o shared artists) We listen to each others personal radios etc. But since after this update he is not my neighbour anymore (OK, I may unerstand it) but few minutes the taste-o-meter started to show me Very Low. You share [b]a few artists[/b] in common. Is over 300 a few? I don't like the new formula...

    9 Nov 2006, 15:40
  • Century_Skunk

    ^same problem for me. Me and my wife play all the same MP3's from the same PC and at last look shared over 250 artists in common. Now we aren't even neighbours and apparently are very low compatibility. Even visually it's obvious from our chart this is a bit screwy

    9 Nov 2006, 19:19
Ver todos (112)
Dejar un comentario. Entra en o regístrate (es gratis).