I also implemented the concept of supervoters, which are a rotating group of users whose favorites are highly correlated to mine. Any posts these users vote on will get +5/-5 in contrast to the normal +1/-1.
So anyone above Janitor, and the users with the "Approval permission", can do +5/-5. Quite a responsability there.
dean_exia said:
[...] But I fear the way score works now becoming the one in Pixiv, high favorites/scores on comics with sub-par quality because people loves the context of them.
I guess, there should be a limit of scoring by copyright or artist, KC or Touhou cannot be scored just by quality (Kashima and Iowa posts are out of control already)
As more people have been signing up, the quality control aspect has consequently decreased.
There's been more than a few reasons quality control took a drop over time, and it's not just the size of the userbase increasing. I don't think folding to that even more with scoring just to try and go with the flow somehow is a good move.
Lannihan said:
So anyone above Janitor, and the users with the "Approval permission", can do +5/-5. Quite a responsability there.
Honestly, I wouldn't like being able to +5 or -5 any post (I don't have this). It wouldn't seem fair if I'm already a Moderator right now, or if I was even a Janitor.
I guess, there should be a limit of scoring by copyright or artist, KC or Touhou cannot be scored just by quality (Kashima and Iowa posts are out of control already)
dean_exia said:
No wonder there's a lot of images with 30+ scores lately. But I fear the way score works now becoming the one in Pixiv, high favorites/scores on comics with sub-par quality because people loves the context of them.
That's a big part of the problem with this. Nothing is going to change a character or copyright from being in the popular posts except other popular characters or copyrights finding their way in. Other things still had enough of a chance. Members' favorites counting towards that again will just make that stick even harder, especially with anything posted just after 12:00 AM EST when the new day starts (from my timezone, anyway, correct me if I'm wrong). There's still Members that will favorite literally everything in sight, and that's part of why their favorites stopped counting towards score in the first place.
Having scores on posts regularly reaching upwards of 50 or even 100 in under one day again after years of having more controlled scoring feels like starting with Soccer and suddenly jumping to Pinball. If score still had some meaning before, this just removes what meaning was left. The site is way past the point of going back to that. It's too much, and it should be reverted.
No wonder there's a lot of images with 30+ scores lately. But I fear the way score works now becoming the one in Pixiv, high favorites/scores on comics with sub-par quality because people loves the context of them.
Well, that and the fact that a lot of the images posted before this change are going to start to be superseded by the scores of newer posts, even though the score identifier meant a lot more before this change as only gold and above members could vote (privileged back in the day), plus it also used to be that favoriting did not automatically add to the vote count so the correlation used to be even lower.
That means, it won't be long before using the "order:score" with other tag combinations will show only newer posts and start pushing older posts down in popularity, even though the older posts' scores were achieved under stricter conditions.
There'll also be less meaning or need for the "order:favcount" metatag, since this change will increase the correlation even higher. I've used the "order:favcount" frequently in the past, because although subtle, there has been a difference between that metatag and the "order:score" metatag.
Apollyon said: That's a big part of the problem with this. Nothing is going to change a character or copyright from being in the popular posts except other popular characters or copyrights finding their way in.
The idea isn't to reduce the popularity of a character in the site, my point is to control an image score instead of having 20 votes at the first five minutes of posted, to 4 or 5, whats is normal (probably). Because the main problem here is the quality related to the score, so being more strict with the most popular copyrights will give more chances to other better works can be reconogized and knowed for most of the community.
So, the restriction of +5/-5 to a limited number of users, and the idea, at least can secure a more objetive scoring I think. If it possible too.
Apollyon said: Other things still had enough of a chance. Members' favorites counting towards that again will just make that stick even harder, especially with anything posted just after 12:00 AM EST when the new day starts (from my timezone, anyway, correct me if I'm wrong). There's still Members that will favorite literally everything in sight, and that's part of why their favorites stopped counting towards score in the first place.
Having scores on posts regularly reaching upwards of 50 or even 100 in under one day again after years of having more controlled scoring feels like starting with Soccer and suddenly jumping to Pinball. If score still had some meaning before, this just removes what meaning was left. The site is way past the point of going back to that. It's too much, and it should be reverted.
I agree, but this is a matter of time. Because still is a temporaly change, I hope.
The idea isn't to reduce the popularity of a character in the site, my point is to control an image score instead of having 20 votes at the first five minutes of posted, to 4 or 5, whats is normal (probably). Because the main problem here is the quality related to the score, so being more strict with the most popular copyrights will give more chances to other better works can be reconogized and knowed for most of the community.
I probably wasn't clear with what I said, but strictness is what I was referring to. Instead of directly clamping down on popular copyrights, maybe we could have more accessible pages to show popular posts in particular copyrights (the people that can tailor their own popular lists with just tag searches, can still do so).
So, the restriction of +5/-5 to a limited number of users, and the idea, at least can secure a more objetive scoring I think. If it possible too.
This in particular I might be willing to test, even if I don't feel like I actually need that ability to tip scales on posts by THAT much.
If the change stays, I agree that scores should be recalculated for older posts. Unfortunately, it's impossible to calculate precisely, since votes are pruned after a few months and only count is stored. So, unless it's possible to temporarily restore all votes from backups, we'll have to come up with some sort of formula - at the very least, older posts shouldn't have upvote count less than fav count.
Also, this essentially gives regular members a limited ability to vote. Should direct voting be allowed, too? Though, just like that change, it will heavily tip the scales towards rating:e. Check out the difference between order:favcount and order:score; the latter will soon cease to exist.
You all should also note the other change: with this commit, gold+ users have lost the ability to give a post +2 score (one with favorite, one with vote) - because favoriting now literally creates a vote, trying to vote for a favorited post will give you an error - "you have already voted". I'm pretty sure concept of supervoters was introduced to balance this change, though +5 feels like overkill.
Also, @albert, is the change not deployed to testbooru? And, if selecting supervoters is going to be completely automated, you should be aware that you're open to an exploit - anyone can create an exact match to your favorite list.
Is it just me, or am I the only one who dislikes effectively giving voting power over to any member with the favourites change? Especially if there's like 100 free members to every gold, platinum, builder/janitor or above? You can get a picture about Iowa or Kashima which squeaked in as being 100+ scored really easily, even if the favorite vote worked in even a fractional manner, much less a 1-to-1 correlation.
I don't favourite everything, but I've seen people who do... and that basically makes the scores worthless for anything but determining most popular "at this time" searches.
Unless we wipe all past scores and start fresh as of a given day, I don't know how to make them meaningful.
Is it just me, or am I the only one who dislikes effectively giving voting power over to any member with the favourites change? Especially if there's like 100 free members to every gold, platinum, builder/janitor or above?
I don't know if it was intentional, but this sounds a lot like whining about losing a privilege of being gold or above instead of criticism about how this is implemented at the moment.
That said I also think there are some problems.
The fact that everybody can now effectively up-vote, but only gold and above can down-vote will skew score a lot towards the positive side. Assume a post that 50% love and 50% hate. With the old system the score would probably have ended up somewhere close to zero. With the new system where everybody can up-vote but only a relatively small number of people can down-vote the same post will have a possibly high positive score.
Old posts will inevitably have a much lower score if the score won't be recalculated on them which restricts the usefulness of ordering by score because older post will have no chance of appearing.
Also some comments above seem to make the assumption that gold users make for some magical reasons better decisions when voting, but I can't see any reason why this should be the case. If we really care about the quality of score I think it would make sense that favouriting doesn't affect score and voting has to be done explicitely, that everybody can up- and down-vote and that the number of votes everyone can cast is in some way limited (this limit could be varied per user similar to the upload limit).
If we really care about the quality of score I think it would make sense that favouriting doesn't affect score and voting has to be done explicitely...
I can't think of a post I have ever favorited that I also did not want to upvote as well, therefore having these tied together makes sense.
SD-DAken said:
...everybody can up- and down-vote and that the number of votes everyone can cast is in some way limited (this limit could be varied per user similar to the upload limit).
However, given your idea, this would make a lot more sense. I think limiting votes could be of some use, that way people would have to be pickier about what they want to upvote, ensuring that users that favorite almost every post of certain tags don't automatically increase the popularity of those tags.
Of course, there are a couple of ways limits could be set:
A flat rate for all users
A progressive rate for user levels
A quality based metric
Something else
There could also be a separate permission like the "approve posts" and "ulimited uploads" permissions, like "unlimited voting".
I don't know if it was intentional, but this sounds a lot like whining about losing a privilege of being gold or above instead of criticism about how this is implemented at the moment.
That said I also think there are some problems.
The fact that everybody can now effectively up-vote, but only gold and above can down-vote will skew score a lot towards the positive side. Assume a post that 50% love and 50% hate. With the old system the score would probably have ended up somewhere close to zero. With the new system where everybody can up-vote but only a relatively small number of people can down-vote the same post will have a possibly high positive score.
Old posts will inevitably have a much lower score if the score won't be recalculated on them which restricts the usefulness of ordering by score because older post will have no chance of appearing.
Also some comments above seem to make the assumption that gold users make for some magical reasons better decisions when voting, but I can't see any reason why this should be the case. If we really care about the quality of score I think it would make sense that favouriting doesn't affect score and voting has to be done explicitely, that everybody can up- and down-vote and that the number of votes everyone can cast is in some way limited (this limit could be varied per user similar to the upload limit).
I think my concern is less about privilege and more about how posts that are about popular characters but which may not be the best examples will get higher votes by being favourites from accounts which may be sockpuppets registered solely for that purpose.
By requiring someone either pay real money or be recognized by Albert as a contributor to the site, I feel it's less likely that we would run into issues with posts that aren't representative of the best examples being upvoted simply because people registered them as favourites.
Maybe if we could start work a clean slate work scores this might work, especially if those designated voters of Albert's have the power to downvote posts that aren't really that good.
I noticed this vote to favorite ratio discrepancy recently with one of my uploads and I was wondering why. Now I know. Also, about that supervoter proposal, I think it's okay if you have like at most ten people to have that permission.
BrokenEagle98 said:
There could also be a separate permission like the "approve posts" and "ulimited uploads" permissions, like "unlimited voting".
A separate permission for unlimited voting? Sounds like a bad idea for something you just had to pay before to get.
edit:Now that I'm looking at more posts with a high vote-to-favorite ratio, I feel like we should go back to the old system. Otherwise, we end up like we did before a favoriting system was implemented or like Gelbooru.
Uh, so that's why some recently uploaded posts had such abnormally inflated score.
Honestly, I can't see anything good coming out of this. Now score is even less indicative of image's quality and is all about the popularity. And being from popular artist, copyright, or fetish gives them way, way more advantage than before.
Besides making score noticeably less meaningful, there are also two objective flaws in the new system: 1) There is nothing that would stop users from creating multiple sock puppet accounts to upvote posts by favoriting them. This was actually one of the best advantages of the old system - it was largely protected from manipulation and abuse. 2) Old posts are at huge disadvantage when their score is compared to that of the new ones. They were already at a disadvantage before due to the naturally increasing number of people who can affect score, but at least it was somewhat limited. Now the gap seems to be an order of magnitude larger.
As for new supervoters, I can't clearly trace their influence since their votes are not indicated in any way, but the concept itself seems counterproductive. When supervoters are selected based on similarity of their tastes to a single person, it will only lead to certain posts getting advantage because they fall into that person's areas of interest. And even if the selection process is changed and range of interests diversified, the bias will still be there. I don't see how making some preferences more influential than others is beneficial for the system as a whole or useful for other users who may not share those preferences.
I think that it would be a better idea to move in the opposite direction. Make the score of recently uploaded posts less inflated, not more. One simple way to do it is to revert all recent changes to the system except favorites counting as upvotes, so that single (vote-capable) user couldn't +2 a post right away. This would help to keep parity with older posts. It would also help to slow down the snowball effect (comparably higher score -> more visibility -> even higher score) since the gap in score between posts would grow slower and various posts would have more time to get noticed.
3 years ago the code where janitors/mods were given +3/-3 votes was taken out because it gave individuals too much power, now new code where random people get +5/-5 was added because ???
I'm gonna have to agree with MyrMindservant and Log here. Why have supervoters when the same privilege was given to mods and janitors years ago and taken away?
I never regarded scores as being particularly useful before so I don't feel like much has been lost. I would rather a personalized system that suggests images favorited by users with similar tastes to you, or simply "users who liked post X also liked post Y".
And the issue of old images being disadvantaged compared to new images exists regardless of what system is used. There ought to be a way to bring undiscovered images (that have a low score due to having low views rather than due to being disliked) to people's attention, as well as a way to bring old but popular images to the attention of new users who haven't seen them yet.
I think my concern is less about privilege and more about how posts that are about popular characters but which may not be the best examples will get higher votes by being favourites from accounts which may be sockpuppets registered solely for that purpose.
If the concern is about sock puppet accounts we could restrict the ability to vote to accounts with a certain minimum age (or some other criterion?).
MyrMindservant said:
Honestly, I can't see anything good coming out of this. Now score is even less indicative of image's quality and is all about the popularity. And being from popular artist, copyright, or fetish gives them way, way more advantage than before.
But images from popular artists or copyrights not only have a higher chance of being favourited by basic members but also by gold users. This means even with the old system the score was likely skewed towards more popular artists and copyrights, it's just the factor of the skew that possibly increased with the new system. I don't think the right thing to do against this is to restrict the ability to vote to gold and above again, but instead separate voting and favouriting (see my previous post).
The supervoters bring imbalance to the voting system, especially because there are only very few of them. Also 5x the voting power is overkill. Normal users getting voting powers enables a more broad range of voters but also creates the possibility to cheat with sockpuppet accounts.
Now we could either get rid of the new voting system or try to resolve the problems, like by giving higher level users more voting powers (but not that high) to level out the field. Something like normal users 1 vote, gold and up (promo incentive) 2 votes, contributors and approvers (both had to qualify for the level) 3 votes.
To prevent easy sockpuppeting normal accounts should get voting power after a longer probation period, like some months. If accounts have no or little activity during that period the probation should start anew. That should be quite effective against users with multiple accounts.
The supervoters bring imbalance to the voting system, especially because there are only very few of them. Also 5x the voting power is overkill. Normal users getting voting powers enables a more broad range of voters but also creates the possibility to cheat with sockpuppet accounts.
Now we could either get rid of the new voting system or try to resolve the problems, like by giving higher level users more voting powers (but not that high) to level out the field. Something like normal users 1 vote, gold and up (promo incentive) 2 votes, contributors and approvers (both had to qualify for the level) 3 votes.
To prevent easy sockpuppeting normal accounts should get voting power after a longer probation period, like some months. If accounts have no or little activity during that period the probation should start anew. That should be quite effective against users with multiple accounts.
If it is possible to do, that's probably the best proposal hitherto if we really want to change something. I still don't get it why a "winning" system should be changed. It worked fine over the past few month or not?
If the concern is about sock puppet accounts we could restrict the ability to vote to accounts with a certain minimum age (or some other criterion?).
Ah, but then do we count for favourites added some months earlier as votes once the accounts they're attached to l reach a certain age? If so, what sort of load does that add to the back end? If not, we've only delayed the issue by that long. I'm not convinced that making favourites count automatically as votes is a good idea, period. I may vote a post I've favourited, but I may also have done or to look at later.
The sock puppet thing still isn't handled by that - not unless there's a way, at registration, to note IP addresses that have all share the same address on registration and then ban them - or some other solution like Schrobby's suggestion.
But even we have to figure out what constitutes activity - uploading pictures? Forum activity?
