Weekly scrobble cap?

  • 4
  • Idea
  • Updated 5 months ago
  • Implemented
Archived and Closed

This conversation is no longer open for comments or replies and is no longer visible to community members. The community moderator provided the following reason for archiving: Archiving because I don't want to draw too much attention to this (in case people use it to figure out what the limit is and how to game system). But know that we have this in place now.

I know this has been brought up around the forums in different threads, but are there actually any plans to curb blatantly false scrobble counts?

I've just stumbled on yet another cheating user who, according to his Last.Week (that excellent tool of yours that easily and quickly reveals who's cheating), listened to 21 days of music in the past 7 days.

Could you at least try and put a crude cap on, something like 3,000 scrobbles per week? I think a solid 7 days of listening is nearer 2,500, but this particular user I've just seen has supposedly scrobbled almost 9,300 times in the last week.

Perhaps you could use the data you're already gathering for your Last.Week tool and freeze a user's account for a month if they've scrobbled over 7 days worth of music; do it three times and have your account deleted? Alternatively, introduce a "Report Profile" flag (requiring a reason for the report)?

The charts and listening data/histories of users is meant to be one of Last FM's distinctive features and it would be nice if you're working on a way to maintain/regain their integrity. What is the point in you introducing badges for artists to show who are the Top 100 most-played if the data they're reliant on is bloated by lots of fake scrobbles?

FWIW, I have started unfollowing users who's Last.Weeks reveal them to be cheating, but it's still something I would like to see addressed by yourselves, as I see it as a site issue for all not just the fact that it irks me.
Photo of Elliot Robinson

Elliot Robinson

  • 885 Posts
  • 2154 Reply Likes

Posted 2 years ago

  • 4
Photo of G.G.B.

G.G.B.

  • 420 Posts
  • 273 Reply Likes
I don't care so much about the top listener ranking, but the real problem with faked scrobbles is that they influence the general music statistics and therefore manipulate the recommendation system in a negative way.  People get less recommendations based on people's taste, instead they get music based on faked scrobbles from users who think this site is a competitive browser game. 

I hope there will be a solution which keeps the stats clean, though it should not prevent users from catching up on missed scrobbles or importing a backup which includes very often many thousands of scrobbles.
Photo of gw666

gw666

  • 397 Posts
  • 207 Reply Likes
Are you sure it influences the general statistics? I thought it was mainly based on the number of users and not the (inflated) number of listens?
Photo of G.G.B.

G.G.B.

  • 420 Posts
  • 273 Reply Likes
The statistics are influenced definitely, because all the scrobbles are counted on the public page, no matter how they were created. The question is, does the personal playcount really have an impact on the internal system / the recommendations?
I don't have that information, but I would be disappointed if it's not the case. I assumed that the algorithm uses all numbers which reflect the taste of people, and that is also the total playcount of special songs. So it can recommend you more popular songs similar to your top scrobbled songs. By excluding the playcount number from the recommendation system they would miss the opportunity to use valuable data and it wouldn't even be enough to keep the system clean, because fake scrobbles are not necessarily repeated. If the recommendation algorithm is smart enough, it could detect possible fake scrobbles and filter them out, or in uncertain cases reduce their weight in balance.
Photo of Jon

Jon, Community & Customer Services

  • 4863 Posts
  • 3607 Reply Likes
>>>does the personal playcount really have an impact on the internal system / the recommendations?

It has negligible impact. Our recommendation algorithm has nearly a dozen filters in place (at various levels) to clean, denoise, normalise, and reduce the influence of users who scrobble a small number of tracks or artists excessively.  It even has filters in place to prevent feedback loops from listening to the radio (otherwise listening to similar artists radio, for example, would reinforce the similar artist scores).  

On the individual user level, we have specific filters in place to model the scrobbling preference of users, weighted against time, to "downweight the influence of users who submit large numbers of scrobbles for a very small number of artists (users with "L-shaped profiles")" and "to avoid the preferences of users with very high numbers of scrobbles dominating similarity scores". Then after calculating the similarity score between two artists, we apply another function which "downweights similarity scores if they result from the preferences of only a few users". But there's much more to it than that (tagging an artist has an impact, for example, so does loving a track, banning track, etc) and even once we've calculated similarity scores, we run them through yet another series of filters to re-rank and reorder them based on tag, reach, time, and other factors.

I can't go into the details of exactly how our similar artists and recommendations are calculated, but hopefully this at least gives you an idea of some of the factors that are considered. If you over-scrobble an artist or track, the only music recommendations it will impact are your own personal recs.
(Edited)
Photo of Elliot Robinson

Elliot Robinson

  • 883 Posts
  • 2153 Reply Likes
This is really interesting, thanks, Jon.
Photo of Jon

Jon, Community & Customer Services

  • 4871 Posts
  • 3623 Reply Likes
Official Response
As I said last week, both our lead developer and head of product are keen to implement a daily scrobble cap at some point in the near-ish future. We're planning to do more api work in the latter half of this year, so I will remind them of this nearer the time.   If we do go ahead with this, understand that it will be a very reasonable cap, and the aim is prevent extreme cheating (using scripts and/or abusing the universal scrobbler site) that causes us genuine problems (millions of scrobbles per month), and not people scrobbling while they're asleep, shared / group accounts, radios/jukeboxes, or "That guy is at the top of my last.week leaderboard AGAIN" kind of cheating (just ignore / unfollow them).   
(Edited)
Photo of James Joul

James Joul

  • 983 Posts
  • 2019 Reply Likes
Thanks. I think it would be unreasonable to expect Last.fm to deal with anything other than the glaringly obvious fake accounts. Not only would it be a huge commitment but it's almost impossible to tell if people are "cheating" on a smaller scale. Like you said, ignore and unfollow. 
Photo of Elliot Robinson

Elliot Robinson

  • 885 Posts
  • 2154 Reply Likes
As I said last week, both our lead developer and head of product are keen to implement a daily scrobble cap at some point in the near-ish future.
This is great news.

...the aim is to prevent extreme cheating that causes us genuine problems , and not people scrobbling while they're asleep, shared / group accounts, radios/jukeboxes, or "That guy is at the top of my last.week leaderboard AGAIN" kind of cheating (just ignore / unfollow them).
That is all I was suggesting - hence the 3,000 scrobble cap (which is a generous overestimate, I would say). Personally, I don't agree with scrobbling while sleeping but I can cheerfully ignore it as at least it's not flying in the face of the space-time continuum, lol - as you suggest: unfollow.

As James mentioned earlier, it's the users that are treating it like a competition that are (in my opinion) the problem. 21 days of scrobbling in a 7 day period are the kinds of users I would expect are the ones causing you an issue, too.

Knowing that you're aware of, and intending to address the issue at some point in the future is good enough for me.

Thanks for replying, Jon.
Photo of G.G.B.

G.G.B.

  • 420 Posts
  • 273 Reply Likes
Just a reminder: Importing a backup with tools like the Lastfm Scrubbler could break the limit, because it can not associate scrobbles to their correct time in the past, when they are older than 2 weeks (Last.fm limitation). So it would be useful to have a working import/export function for the scrobbles before installing a scrobble limit which prevents backups.
Photo of Jon

Jon, Community & Customer Services

  • 4871 Posts
  • 3623 Reply Likes
This is now implemented.

This conversation is no longer open for comments or replies.