Donmai

Danbooru 2 Issues Topic

Posted under General

This topic has been locked.

Type-kun said:

What I dislike about this change is that score thresholds changed unevenly. Previously, I knew what quality to expect on average when searching for scores like 5+, 10+, etc. Now, new posts that would formerly be 5 are usually over 15, while old posts from same group stay at 5.

So is there anyone that actually sees a use or positive effect of this score acceleration, the inflated score scale and the other effects of this change? I think these can be reversed now. It's been over a week.

Updated

The goal was to have less of a correlation between favcount and score correct? We'll have to see if albert saw a decrease in correlation between the two since the implementation of the new features.

reese said:

The goal was to have less of a correlation between favcount and score correct? We'll have to see if albert saw a decrease in correlation between the two since the implementation of the new features.

I very seriously doubt that this was the intended effect. Even a quick consideration should reveal that increasing favorites' influence on score would not decrease correlation between them.

Apollyon said:

So is there anyone that actually sees a use or positive effect of this score acceleration, the inflated score scale and the other effects of this change? I think these can be reversed now. It's been over a week.

I can't say for everyone, but posts I did see in this topic were all against this recent change.

I want to expand on my comment from two pages back. Here is a quote from it:
"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."

Back in forum #112325 I also wrote that it would be better 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". But now, after seeing the practical effects of this experiment and thinking about it some more, I understand that it wasn't the best solution.

The best way to make score more meaningful and useful is to separate it from favorites completely. Upvoting a post should be a conscious choice, not a by-product of favorites.

Currently score mostly depends on favcount, but even before it was heavily influenced by it. This was also the initial issue mentioned by albert in forum #112300. If favcount has too much of an influence on score and score started to lose its meaning/purpose, then separating them seems like a logical solution.
This would also bring some benefits even compared to the previous system:
1) Balance of upvotes and downvotes should be much better when each vote-capable user can +1 or -1, not +2 or -1.
2) Natural score inflation should be significantly slower which would help to keep parity between old and newly uploaded posts.
3) We would get proper indicators for both popularity and score, rather than two indicators for popularity like now or popularity and popularity+score like before.

Updated

Kikimaru said:

Fuck.

I use these to visit artists' sites, and occasionally add their new sites/aliases.
Not happy with this obscuring of information.

Me too.
Builder+ level users should be allowed to see the information. Not to mention that we're knowledgeable enough to not posts these banned artist' works anymore.

MyrMindservant said:
Currently score mostly depends on favcount, but even before it was heavily influenced by it. This was also the initial issue mentioned by albert in forum #112300. If favcount has too much of an influence on score and score started to lose its meaning/purpose, then separating them seems like a logical solution.

That sounds pretty reasonable. +1.

Of course, there's always the issue of how you implement something like this so that there is no significant advantage or disadvantage given to a post based on its age.

Fred1515 said:

Of course, there's always the issue of how you implement something like this so that there is no significant advantage or disadvantage given to a post based on its age.

There is no way to avoid this stuff completely, if only due to the natural score inflation (result of growing userbase).

Currently new posts have enormous advantage, to the point where scores are not even comparable with the ones before the change.
Under previous system newer posts had noticeable advantage, but they were still within the same league.
Suggested system should be mostly compatible with the previous one. For a short while after the change new posts would suffer some disadvantage, but it will fade away after users adjust their habits and start to upvote more. Gradual score inflation will also play a role.

Regarding score inflation, suggested system should be much better than what we had before. Score increase from single vote-capable user being limited to +1, rather than +2, will slow it down. And better balance between upvotes and downvotes will also help.

Regarding implementation, it should be really painless. Score gained from favorites during previous system was already converted into up-score as part of recent change, so old posts won't lose anything. Score of recently uploaded posts can be recalculated to make it compatible with the new system. And the proposed "new" system itself is in fact nothing new. It's mostly the same as previous system, except favorites from Gold+ users won't give +1 to score anymore.

I don't see a point in any of the changes to the score system.

As said by others, the favorites counting towards score is both exploitable and turns score into little more than a second fav count.

Also having these "super voters" drastically increases the personal bias given to both specific users and subject matters. There certainly is a drop in quality happening from time to time, but giving favoritism to certain users to inflate score counts isn't going to help that.

People that upkeep the site either by donating or contributing should be the ones getting the ability to vote on score, but when certain people get preferential treatment over others and anyone's favs boost score it undermines the entire system.

Jigsy said:

Not entirely sure if this a bug or not, but is there any reason why blacklisted tags are always visible by default? (At least on my end.)

Did you click "Disable all?" The only way to long-term re-enable them afterwards is to click "Re-enable all." Re-enabling them individually won't stick between pages.

tapnek said:

He thinks there may not be a change in correlation and given that some of us are annoyed by this feature, I think he should revert it.

A little late to the party with my imput, but a change in correlation seems somewhat obvious from my perspective, so I'm seconding this.

The easiest way to look at it for me at least wound up comparing the scores from December, January, February, and March.

In December there is one post that got to a score of over 100, in January, there are three. In February, when this change was implemented, there are twenty-three. March is barely over a third done, and so far there are thirty-eight images with scores of over 100.

If that isn't a drastic difference, then I don't know what is. If the goal of this change was to uncouple favcount from score, then it seems to have had the opposite effect while also serving to dilute the inherent meaning of the score that was built to this point.

Updated

Grahf said:

If that isn't a drastic difference, then I don't know what is. If the goal of this change was to uncouple favcount from score, then it seems to have had the opposite effect while also serving to dilute the inherent meaning of the score that was built to this point.

Highlighting this particular point. It's two steps back, removing the meaning that score had (in spite of people that don't think it mattered), while opening up more problems involved with score.

In addition to all this, apparently there are some users who are granted supervoter privileges without them even knowing or there being any indication that they are supervoters in the first place. This doesn't seem to include approvers or mods.

Provence said:

It seems that the scoring system got reverted?
Just clicked through some posts and the score is on some posts highly below the favorite amount.

Confirmed... it went into effect 3 hours ago.

To sum up, only Gold+ and Supervoters can vote now. Gold+ users only get +1 for the vote count (still no more +2), and Supervoters get +3 for the vote count. Admins are Supervoters by default.

Provence said:

It seems that the scoring system got reverted?
Just clicked through some posts and the score is on some posts highly below the favorite amount.

Scoring seems to have been reverted in some parts of upcoming uploads, but older posts under that system still have favorites from all members counting towards counting towards their score.

It would make sense to recalculate the scores on those posts from when this test was implemented to the standard scoring before any changes happened. Assuming this isn't already planned.

Yeah, the results didn't seem worse but they also didn't seem better so I decided to revert it.

Unfortunately fixing the scores is intensive so I'll probably just leave them as they are. So the month of March will have some abnormal scores.