Danbooru

Question: Uploading with minimal tags.

Posted under General

Those pixiv stalkers need to set up their own forum in which they decide uploading schedules. This would save all of them (and us?) a lot of time. Although I wish they would all tag as well as Schrobby does.
Day 1: A uploads, B to Z focus on tagging.

S1eth said:
Those pixiv stalkers need to set up their own forum in which they decide uploading schedules. This would save all of them (and us?) a lot of time. Although I wish they would all tag as well as Schrobby does.
Day 1: A uploads, B to Z focus on tagging.

Communism never worked out very well for the Soviet Union.

Kikimaru said:
I still have no damn idea how to get Contributor status, and I'm more active than most of you. >_>

This is not meant to be a slight or offensive in any way, but the problem is likely the fact that close to a fifth of what you posted was deleted for one reason or another. The few times criteria for inviting people past the mod queue has been discussed (long before I became a moderator), one of the chief criteria was that they have a very low proportion (and ideally low count) of deletions.

The risk of ignoring that criteria is that someone get a free pass to push whatever that proportion of posts that wouldn't have otherwise been accepted into the system.

What should ideally be implemented (and currently isn't) would be a way to do a weighted sample of a users' posts towards recent history. That way it would be possible for users that had a lot of early deletions to be seen in a preferable (and more accurate light) after they learn what content is good and well received. That way someone can be invited with a reasonably good expectation that their current good performance will continue rather than the fear that early poor performance will resurface.

The current system makes it difficult to make this sort of judgement, and subsequently the increases the risk that a mod or admin will put their name on the line for someone who may or may not prove to be posting acceptable stuff in the future.

People have been reverted in the past when invited hastily, and it's much harder and more drama provoking to clean up approved posts that shouldn't have been, than to screen them out in the queue (notice what happened when Hazuki or Rantuyetmai for example have tried in the past).

Unfortunately, this places well established users with many deletions in a difficult spot, because it's often very hard to reduce that proportion or otherwise prove that their recent postings are mostly all getting accepted with the current system. New posters with 100 posts, all approved currently look preferable to an established user with 1000 posts, 200 deleted, even if the last 600 all were accepted.

Updated

Shinjidude said:
People have been reverted in the past when invited hastily, and it's much harder and more drama provoking to clean up approved posts that shouldn't have been, than to screen them out in the queue (notice what happens when Hazuki or Rantuyetmai for example have tried in the past).

Rantuyetmai got into shit because he kept deleting duplicates and revisions behind our backs, and despite at least two forum threads complaining about his actions and an explicit warning by jxh2154, the first thing he did as a newly-promoted Mod was to make his deletions permanent to better hide his tracks. It had nothing to do with cleaning up bad-quality posts.

Action_Kamen said:
Don't you LOVE it when you find and start uploading some obscure artist on pixiv or a blog; a cool artist that you like who illustrates stuff that meets danbooru's quality; only to log on some time later and find someone stalking him. Then you're forced to find another dude then the process repeats itself.

I love it.

I have uploaded (not this account) the first image (to danbooru) of an artist before and then run into the stalking you are talking about.

It used to bother me but now I look at it as a point of pride that I introduced an artist that is worth uploading to danbooru.

Updated

Kikimaru said:
I would also like to add that 63 of those deletions happened after the artists elected to not have their art on Danbooru.

I thought all images like ones that the artist asked to be removed were double deleted so as to not affect the ratio?

Assuming I am recalling one of the functions of double deletion.

Pyrolight said:
I thought all images like ones that the artist asked to be removed were double deleted so as to not affect the ratio?

Assuming I am recalling one of the functions of double deletion.

Users whose upload rates are massively affected by artist takedown requests can ask to have their upload cap manually changed but the images are not double-deleted.

Fred1515 said:
IMHO the biggest part of the problem is simply the number of active uploaders. I don't know if Danbooru's user base has grown so much that pixiv and similar sites cannot produce enough quality art to keep up (personally I doubt it) but since the number of active uploaders has increased it's only natural that getting promoted takes more time than it used to.

Yes that's true. There's more art produced than in the past but even more uploaders. And don't get me wrong, that's a good thing, since this way we don't have to worry about new stuff being missed.

There are more reasons as to why promotions take more time though:
- requirements used to be too easy, sometimes leading to lots of bad uploads and then demotions like Shinjidude said,
- the userbase has grown while the staff hasn't, so the workload to monitor who does well has become bigger,
- let's be honest, few elder users actively give a fuck about promoting people, perhaps expecting someone else will. From following the various promotions given since 2010 it's something like 1/3 albert's periodic checks, 1/3 me dmailing Hazuki or jxh, and 1/3 other users, generally from records.

And I forgot one more thing earlier about getting noticed:

User records are effective at helping promotions.
Now, when you look who gets records from well advised users (Janitors and such, because otherwise the records are sometimes kinda absurd) for their uploads, it's people who consistently post the highest quality content of the site, and it's no secret that pretty much all of it comes from the surface of Pixiv and Imouto/Yande.re.
So even though your upload history is such only because you happen to be tad faster or better organized than most of your fellow racers you're still the one earning the rec.

Uploaders whose history is good but falls short of eye candy don't get recs, and so they have a harder time getting promoted.
I don't know if it's because they don't catch enough attention, because they don't get the viewer fired up enough for a record, or because they look like they would be unable to tell what top quality is.

saizo0070 said:
Hell I'm still wondering how I got mine's in the first place. I'm not complaining but still.

The day you were promoted, albert also promoted 7 other Contributors. You must have been recommended by someone or been lucky because when I looked at those 8 profiles at this time, 3 of them including yours weren't meeting the usual requirements for Contrib.

Shinjidude said:
Unfortunately, this places well established users with many deletions in a difficult spot, because it's often very hard to reduce that proportion or otherwise prove that their recent postings are mostly all getting accepted with the current system. New posters with 100 posts, all approved currently look preferable to an established user with 1000 posts, 200 deleted, even if the last 600 all were accepted.

There aren't that many big posters with a high deletion ratio so it's easy to keep track of them. Rather, the sad part is that they don't ever seem to improve even when helped to.

Updated

ShadowbladeEdge said:
You could always do date searches Shinji. Observe Kikimaru's posts in the last 3 months user:Kikimaru date:2012-03-08..2012-06-08. Still about 20% deletion rate FYI.

I know it's possible to do about any sort of search if you query each user individually, but it's much easier (unless you just happen to stumble upon a really good user) to be able to screen with the variables in the user listing. For a hypothetical user with previous deletions it is currently impossible to tell how those deletions are distributed.

Cyberia-Mix said:
There aren't that many big posters with a high deletion ratio so it's easy to keep track of them. Rather, the sad part is that they don't ever seem to improve even when helped to.

True, they do float to the top of the listings if you filter by rank and sort by number of posts. This in itself can be a red flag if they've been passed over by everyone in the past. I still maintain that such users could potentially change their posting habits, and even if they did, it wouldn't be readily apparent from the statistics.

I've never bothered trying to stalk pixiv users. I currently have 526 pixiv links (artists and artist responses) queued up in this script. Pick a random link, enable EPP's source search to find posts that haven't been uploaded, pick out anything you like, use PTP with the Danbooru mods to tag uploads faster, clear artist, repeat. Add artists of posts you fav, the artists' last 3 favs, linearts they colored, etc... It doesn't end.

Shinjidude said:
I know it's possible to do about any sort of search if you query each user individually, but it's much easier (unless you just happen to stumble upon a really good user) to be able to screen with the variables in the user listing. For a hypothetical user with previous deletions it is currently impossible to tell how those deletions are distributed.

Like, X% of the user's most recent Y posts are active? That would be easy to add to the Miscellaneous script if that's a feature you really want (and you use that script).

...

</plug>

Updated

Shinjidude said:
I do use and love the Miscellaneous script.

Glad to hear it. :)

Shinjidude said:
though it'd still only show up in user profiles or direct queries in all likelihood, right? Still useful though.

Where else ought it show up?

Shinjidude said:
Maybe I should work on scripting it in

No need to make the effort yourself if I'm going to be adding myself in a week or so. The hard part is just deciding what exactly it should do and what the default settings should be. Look back at last X posts or posts within X days? For what X?

Updated

I'll just go ahead with the cold-hearted answer:

I'm concerned with the collective quality of danbooru art, not the individual enjoyment of a given uploader.

If we get a million great images from a hundred people or a million great images from a million people, it's the same outcome. Especially for the 99%+ of Danbooru users who only browse, not upload.

I am actually sympathetic to Cyberia's points. I got frustrated back when I uploaded regularly too. It's sort of part of why I almost stopped, but honestly 90% of the reason is admin work and just having minimal free time in life in general.

But I don't think there is a reasonable, easily implementable solution, nor do I think the problem warrants the work that would be required to implement a very complicated solution.

Will some quality contributors get upset and eventually leave? Yes.

Will a dozen other eager users upload the same exact images that first contributor was going to upload? Yes.

Will I as a viewer of an image care why uploaded it? No.

So educate bad taggers. If they stay bad, neg them. If they are just egregiously bad, they might get banned. But the whole debate about getting annoyed at getting beaten just does not have a solution. Sorry to put it bluntly but any other answer would be dishonest.

RaisingK said:
Where else ought it show up?

I was initially thinking the user listing page (along with the ability to sort on that variable). To do that with a userscript (especially the sorting bit) would likely mean a bunch of hammering the DB for pages of data and Ajax-fu to sort it all out, and would probably just be impractical. The user listing often times out even with the queries it's built to run by default.

It might be feasible to just load the metric alongside the user though as the script does for post scores (though even there each hit would be slower and harder on the DB than a single score lookup in this case, in all likelihood).

RaisingK said:
The hard part is just deciding what exactly it should do and what the default settings should be. Look back at last X posts or posts within X days? For what X?

This is sort of a judgement call. Calculating for the last X posts (maybe 500) would be best to ensure a uniform sample size, but may not be sufficient for extremely prevalent posters who might have had a couple of good weeks, but going back by a month shows lots of deleted posts.

Going by time (maybe a year) might be better to ensure any changes in posting quality are persistent, but could cause problems with users with low post counts over that period). A person who posts only twice in the last year might have a 100% approval rating for that period, but not have provided enough data to draw a conclusion from.

In any case, and as would normally be the case, closer inspection would reveal potential issues, but having a well defined metric would provide a good goal for existing users, help more easily winnow out users worth taking a second look at, and be a good thing to point to when justifying user invites, especially if they had a shakier past.

What are others' thoughts on this? Is it worth the effort?

Updated

This could be useful.
Just a random idea though, would having full-access to the DB be useful in anyway? I've got an API scraped DB I've been using for my pixiv script which I'm guessing could be better to use than doing x amount of api calls.

Also wouldn't it be better to have something like this as a separate site/page rather than a userscript? Possibly using some kind of chart software. (Like one of these)

1 2 3 4 5 6 7 13