neighbours and compatibility scores

  • 0
  • 2
  • Question
  • Updated 12 months ago
  • (Edited)
I have not been following Last.FM for some time, so was only marginally aware of the return of neighbours - thank you, by the way!

Looking at this feature tonight, I see issues with the compatibility scores.(Things like really weird inclusions of artist names. Mr.Kitty seems to rank high in the neighbour matches, which surprises me, given the rank that artist has in my lists. My last.fm name is snyde1)  I recall that they were mysterious in the old Last.FM as well.

I realise that you can't disclose the exact algorithm for compatibilty scores, and see why in some of the search results for this question. However, could you tell us what some of the components are in general terms - how long a period of scrobbling do you look at; are there minimum values (absolute or percent) of scobbles for inclusion of artists; caps for artist scrobbles; order of artists in the two lists; ...

I'm not looking for the vales that are used in the categories (other than period), but a list of the categories used to create the match.
Photo of snyde1

snyde1

  • 108 Posts
  • 71 Reply Likes

Posted 12 months ago

  • 0
  • 2
Photo of Tom Stephen

Tom Stephen

  • 430 Posts
  • 497 Reply Likes
I remember Jon one said, that artists used to "neighbour" system are taken from the time period chosen by you on profile - so, if you use "top week artists", your neighbours will be collected from this list. That's all I know.
Photo of G.G.B.

G.G.B.

  • 415 Posts
  • 272 Reply Likes
Interesting question, by the way, you can calculate the musical compatibility of two Last.fm members using the programme Last.fm Extra Stats. There is also an explanation of the algorithm. Could be interesting to compare the results.
Photo of snyde1

snyde1

  • 108 Posts
  • 71 Reply Likes
Thank you. It is interesting to see that "Extra Stats" is still being supported by C26000. Good to hear!

Back in the day, I had a similar metric in a userscript. (Actually, it looks to be the same, more or less.) A later version also displayed the last.fm match. (I don't recall how good the agreement was - my method was not the same as theirs.)
(Edited)
Photo of AuH2O .

AuH2O .

  • 25 Posts
  • 24 Reply Likes
Whatever algorithm is used for the compatibility scores, I don't think it works really well. I have 15,000 scrobbles and "all time" artists displayed on my profile, and "neighbors" with whom I supposedly have 95% compability can have a top 20 all-time list of artists where I haven't listened to and don't like a single one.

Edit: Just found the perfect example of how the compability score is basically just hit-and-miss: A neighbor with 99% compability (highest I've ever seen). But when I look at their top 50 all-time artists, I have only scrobbled 3 of them. I like none of the other 47 artists, and some I downright hate and wouldn't be caught dead listening to. Justin Bieber? Justin Timberlake?? FFS! There is no way I am 99% compatible with this user. The only thing we have in common is that their all-time top artist is my #5 all-time top artist. The rest is completely different and indeed diametrically opposite, so 99% compability my a**!

As for my compability with the creator of this topic, it is 2%, which conversely is the lowest I have ever seen. But that seems about right, because of his top 50 I've never even heard of 48 of them.
(Edited)