[Deployed] Bot that automatically submits implication requests for costume tags

Posted under Bugs & Features

Fenton

Login_to_view said:

This post is going to be brainstorming more than else, so forgive me if it's a bit of a chaotic mess.

I think this would be very useful. With it's downsides for sure, but automatic alt character wikis would be a godsend, especially for things like Fire Emblem Heroes, which has hundreds of alts.

For the template suggestion, I dislike a single template for all games. I think it would be better to make a master thread that proposes templates (plus an attached placeholder BUR so that people may move onto it - something like update fire_emblem_heroes -> check_post_body, so that users browsing the tag see the little "this tag is being discussed in forum #XYZ").
For example, many FEH Alt tag wikis follow this template:

As for the multi-layered characters, as you said, they can be submitted separately.
(I would like to hear @Fenton's opinion since they're the one currently managing the FE implications.)

I also think that running it once a day is kind of overkill. Maybe once every two days to 1 week? It would also help give some time for fanart of upcoming skins to spawn, so that they may be all implied in a single BUR.

Alternatively, give Builder+ the ability to let the bot run on command?
Something like:
-Ping bot in thread
-Bot checks for thread name, which was previously manually matched with a copyright BUR (eg. the FE thread runs the search for Fire Emblem characters, and uses the FE wiki template)
-Bot checks for tags with certain conditions
-If these tags exist, bot compiles BUR (and wiki if needed), then pushes it in the thread

This would help avoid having 4 back to back 1-or-2-tag BURs in the following days of an announcement.

Another option to avoid that would be to make the bot edit an existing BUR, For example, if the implication is simple (not nested), and the last BUR by the bot is not older than 1 week, the bot will update its own BUR with the new tags.

I personally support this bot. However, I personally think the bot should run if the tag have at least 5 or 10 arts. If they have less than these amount, they would be manually submitted.

Reply
WRS

I'll echo a few things I brought up elsewhere: in response to an edge case NNT found while working on the script, I mentioned that I was having some second thoughts about my suggestion to autogenerate wikis because I didn't really think through all the weird cases it could run into. He mentioned that people might just not create wikis and ignore forum posts.

I feel that most people who are mass gardening for costumes generally aren't writing wikis either except for dedicated costume tag maintainers such as the ones paged to the topic. I've had two instances of people on my own managed copyrights tagging costumes but not writing wikis or making implications which I wouldn't have found out without EventListener or checking newly created tags.

In both the case of generating a skeleton wiki or not, the important consideration I feel is: where are the people to care and/or know enough about the copyright to create/make an edit that's sufficient enough to pass the bot threshold? For gachas I feel that'll be naturally well attended to but for other copyrights I'm less confident that anyone's spending any time keeping up with a topic and its posts to review needed changes or even vote on BURs. We'd probably still not have people doing the needed wiki work just as it is now.

That all being said; ignoring the piece on people caring enough, a skeleton wiki that has an edit threshold I realise may be a good idea if we want to start enforcing a standard or minimum amount of included information such as a pictured example. Early on when I was less familiar with BURs, wikis and all that, one of the pushbacks I got from admins was either shit information (couldn't be understood by the average person) or no picture examples that made review harder. A skeleton wiki is good as a standard setter, but I'm not sure if it'll affect culture surrounding supplying proper wikis for costume tags.

I generally tend to like automated helpers just to make life easier on things that are otherwise taxing to do manually but it's just that one extra part that's missing between making it effective and not being all that different from how people are doing it manually right now. At the very least, one of the features I find helpful already is that it would discover costume tags that aren't implied. That's my biggest issue right now with copyright cleanup, keeping on top of any new tags and making sure that people don't submit wrong implications.

Reply
nonamethanks

The script seems stable enough that I'm going to be able to enable it on the site for a bunch of the series mentioned in forum #350741 within next week at most.

Two problems:

First of all, some series have A METRIC SHITTON of tags without wikis. Fire emblem for example has around 300 of them. For this reason I'm going to skip creating wikis automatically, because this amount of wikis just risks leaving most of them half-finished.
I don't have a good solution for this other than to consider these wikiless tags as "unposted" in their respective topic after a few weeks or a month have passed, so that they periodically get reposted and hopefully are eventually tackled by someone.
It would look like this (using Fire Emblem as an example):

Sample automatic bot post

beep boop. I was going to submit an implication request for these tags, but they have no wiki page.
Write a wiki page for them and I'll be able to do it next time I run.

Tags without a wiki

Self-updating links to all tags without wiki from this series:

Link #1
Link #2
Link #3
Link #4

Second problem: some series don't have consistent qualifiers (like Genshin Impact), and some series don't use them at all (like Cookie Run).

This can be mitigated in two ways:

Both methods have downsides:

  • the first requires those wiki pages to actually be updated
  • the second only catches the top 1000 tags, and in case of big copyrights like Fate that doesn't actually mean that every character will be caught. It also risks catching false positives in case of series where a crossover character generates a lot of overlap (for example: saber (honkai: star rail) (fate)).

Both methods also require a lot more api calls or to go through BigQuery rather than danbooru's api, so this whole thing is a lot more complex on the programming side.

The perfect solution would be to get every single post on the site under, for example, genshin_impact copytags:1, and look at the unique chartags, but that's no viable solution at all due to the sheer amount of posts. That example alone has 230k posts. It would've been doable with BigQuery, but it's not due to issue #4954 (which is likely not going to be fixed any time soon).
I will probably not handle these copyrights in the first trial of the bot, but it'll be something I'll add later, because I have to think on how to do it properly.

Updated by nonamethanks

Reply
Ylimegirl

Yeah earlier I stumbled upon some pre-existing mildly unhelpful wikis for Girls' Frontline skins specifically, like this one. In an ideal world there would be some sort of metatag we could add to wikis to let people know "this wiki isnt finished" (a la banners and maintenance categories on mediawiki projects) but I imagine this is unfeasible. In general I think we just need more people mentally ill enough to write barebones wikis for each skin or at least be able to copypaste the info from another wiki site.

For those curious, you can see which chartags lack wikis with this sort of search, which you can then specify via wildcards and series qualifier/charname to help find the ones potentially relevant to your specific niche. Unfortunately this doesn't help find shitwikis that are just alternate names, like Takeda Shingen (fate) was until I slapped some words on it today.

Updated by Ylimegirl

Reply
Damian0358

Ylimegirl said:

Yeah earlier I stumbled upon some pre-existing mildly unhelpful wikis for Girls' Frontline skins specifically, like this one. In an ideal world there would be some sort of metatag we could add to wikis to let people know "this wiki isnt finished" (a la banners and maintenance categories on mediawiki projects) but I imagine this is unfeasible. In general I think we just need more people mentally ill enough to write barebones wikis for each skin or at least be able to copypaste the info from another wiki site.

We've been slowly chipping away at those ever since NNT yelled at us GFL taggers about it. baconmeh2's been very slowly writing descriptions for some of the wikis, while I've been adding h4. Appearance !post # to others, since assets have been added in the meantime and those seem to be considered a bare minimum substitute in situations like this. They're kind of our burden to clean up, since bacon was the one that added the "[Skin description to be added here]" and I was the one who suggested he do so...

Reply
nonamethanks

Ylimegirl said:

Yeah earlier I stumbled upon some pre-existing mildly unhelpful wikis for Girls' Frontline skins specifically, like this one. In an ideal world there would be some sort of metatag we could add to wikis to let people know "this wiki isnt finished" (a la banners and maintenance categories on mediawiki projects) but I imagine this is unfeasible. In general I think we just need more people mentally ill enough to write barebones wikis for each skin or at least be able to copypaste the info from another wiki site.

For those curious, you can see which chartags lack wikis with this sort of search, which you can then specify via wildcards and series qualifier/charname to help find the ones potentially relevant to your specific niche. Unfortunately this doesn't help find shitwikis that are just alternate names, like Takeda Shingen (fate) was until I slapped some words on it today.

@Ylimegirl you can swap search[has_wiki_page]=no with search[wiki_page][body_regex]=^\s*$ to find tags with a wiki but an empty wiki body. Here's an example for fate. You can also use this trick on the wiki search.

Reply
nonamethanks

I'm going to start up the bot now for a few series. I'll run it manually once a day for a week or so to make sure nothing bad happens (unsupervised automation is bad).
I've decided to submit individual BURs because I figured it was too much of a pain in the ass to vote on BURs that were not editable, and I would've been the one who had to edit them and keep track of the various disagreements for each BUR. It's just much easier to keep track of individual tallies.

Reply
WRS

Yeah, that sounds completely reasonable, and that'd be a whole lot more work which would defeat what the bot's going for if anyone had to resubmit or edit anything besides very specific cases. Sounds good, hope to see a successful trial.

Reply
nonamethanks

forum #357325, I opted to just put the wikiless tags under each BUR so that it creates less forum posts, and so that it only posts them when there's a new BUR.
I really didn't like the idea of repeatedly nagging users to create wikis, and having the bot post without BURs would just be annoying for people who are following specific topics, so if you want to see which tags are still without wiki you'll just have to check the latest post from the bot in each topic.

Reply
1 2 3