Unnecessary blanket tags. Better to keep these separate. They'll have to be backfilled from their respective _neckwear tags and then implied to their base tag.
Even though I was in favor of these tags back then, in restrospect I think it was a mistake to blanket merge them all together. It dilutes tagging and makes it hard to find specific things. What makes sense for eyewear does not necessarily work for neckwear or other types of *wear. We even had someone recently suggest generic topwear tags, which I think is a very bad slippery slope.
Now for some reason we even have a non-descript neckwear tag with 1.3k posts, because people have become too used to the generic thing.
As for reverting this change, it's fairly easy to revert the old BUR by checking the post versions, for example this for red_necktie -> red_neckwear. There's still 4 years of content to sort through but nothing that a mass update or two can't slim down significantly.
I'm ambivalent about these, as I mostly dislike these umbrella garment tags, but they did/do have a purpose. If we do this, we should also really reexamine and plan to act similarly on the other umbrella tagsets too (footwear, legwear, eyewear, etc) for sanity and consistency.
Pros to reverting the umbrella tags: (approving this BUR):
On the tagging and data side, rolling back these aliases open us back up to allow for better granularity and more detail, which I really like.
We could once again discriminate images where tags "overlap" (e.g. a search for blue_necktie can now avoid a picture with both a red_necktie and a blue_bowtie).
Conceivably with knowledge of the current active tags and enough allowable query complexity you an replicate anything we have today with the umbrella tags as far as queries and searches go, and more-so, because we now have that better granularity again.
We stop the trend of more and more awkward, convoluted, and generic tags elsewhere e.g. undone_neckwear
Cons to reverting the umbrella tags:
We lose the ability to easily search by category: (sometimes neck_ribbons and bowties are close enough that you'd like to group them together).
It fragments things such that individual tags are now very specific and only return a smaller number of posts or each individual tag searched.
Maybe as a user I don't know exactly what an ascot is, but would like to find a character wearing one of a certain color.
It will add a ton more tags back to autocomplete (especially if we similarly undo this sort of change to other umbrella tagset categories). Now a "black_" prefix will have a dozen plus tags, and the one you need may get pushed out of what pops up without a longer prefix
We're going to have some inertial issues with taggers. They'll keep using *_neckwear out of habit, and we have no easy way to automate cleaning that up.
Taggers will also ideally need to know and use the most detailed tag available instead of being able to rely on a category. "I know it's red_neckwear, but what exactly is this?" sort of thing.
Even if the past BURs are undoable, there will still be a large amount of manual cleanup involved in undoing these to re-add specificity to posts created during the current system.
It might (arguably should) lead to us undoing other umbrella garment tags
The "right" way
I think the "right" way for us to be able to handle this would be to have some sort of richer tagging system that allows tags to apply to other tags, and "belong" to other tags. Basically turning things into a full-on ontology system like WordNet or the like. That is to say I could tag an image something like "necktie{red, loose}" (just making up syntax here) with the tag "necktie" belonging to "neckwear" belonging to "accessory" belonging to "clothing" (or however detailed we wanted to go.
With that sort of system:
My "necktie{red, loose}" tag could show up in a search for "necktie{red}", "neckwear{red, loose}", "accessory{red}", "necktie", "clothing{red}" etc. allowing your search to be as general or specific as you need it.
There would be no need for umbrella or combo tags to directly apply to a post and inflate the tags there. "red_neckwear", "neckwear", "accessory", "clothing" wouldn't need to be applied to any post directly, but we'd still have the benefit of them acting like combos and umbrellas
A "bowtie{red}" tag wouldn't pollute a search for "necktie{red}".
A "necktie{red}" and a "bowtie{red}" could be tagged alongside each other and be searched for specifically (which you can't do with the umbrella system).
we could conceivably go further and assign that "necktie{red}" to a character and say search for images of "midoriya_izuku{necktie{red}}" and get posts of just him wearing a red_necktie (not others doing so).
maybe "midoriya_izuku" belongs to "male", and then we can search for "male{necktie{red}}" and get all boys wearing red neckties
I'm not sure what that might look like exactly, or how or if it could be made to be performant and user friendly (probably a nightmare), but that's been a pipe dream of mine for Danbooru since fairly early days. It'd be a super complex complete overhaul of the core of Danbooru though, and not something that might happen in the forseeable future if ever.
That all said, I'm not sure how I fall with respect to the idea of us undoing this umbrella system now. I was the opposite of NNT, where I remember being mostly opposed to the umbrella tag paradigm when we decided to go that path, but now question whether or not if it's worth the work and trade-offs to undo it.
We lose the ability to easily search by category: (sometimes neck_ribbons and bowties are close enough that you'd like to group them together).
We have wildcard searches, so *_ascot would let you search for the entire category.
It fragments things such that individual tags are now very specific and only return a smaller number of posts or each individual tag searched.
I don't see why this is necessarily a con; having smaller, more specific tags can make it easier to find what you want than having broad tags.
Maybe as a user I don't know exactly what an ascot is, but would like to find a character wearing one of a certain color.
If you don't know what it is you'd have to look through the entire neckwear tags to find one anyway. We could just as well have a list of neckwear meta-wiki that could help users find them.
It will add a ton more tags back to autocomplete (especially if we similarly undo this sort of change to other umbrella tagset categories). Now a "black_" prefix will have a dozen plus tags, and the one you need may get pushed out of what pops up without a longer prefix
This likely won't be much of a problem since it's unlikely any of the individual tags will be large enough to be at the top of `black_`, and you'll have to type a couple more characters anyway. It might mean certain autocompletes people have muscle memory for now will have to be re-learned, but I don't think it's the end of the world.
We're going to have some inertial issues with taggers. They'll keep using *_neckwear out of habit, and we have no easy way to automate cleaning that up.
This is an issue with essentially every major tag change. We'll just have to hope people will get used to the new changes, sending DMails as necessary. It's also not the end of the world since people can do cleanup of the *_neckwear tags on occasion—no information is lost there.
Taggers will also ideally need to know and use the most detailed tag available instead of being able to rely on a category. "I know it's red_neckwear, but what exactly is this?" sort of thing.
Also happens with current tags. There are plenty of types of clothing I and others likely do not know about. We have clothing request, we could keep current *_neckwear tags as sort of "tagme"s for neckwear, have a separate neckwear_request tag... there are options to deal with this as we have with other similar issues.
Even if the past BURs are undoable, there will still be a large amount of manual cleanup involved in undoing these to re-add specificity to posts created during the current system.
It might (arguably should) lead to us undoing other umbrella garment tags
The latter point is exactly why I started out with neckwear, since it will probably be the easiest to cleanup. No point in me writing up a dozen BURs if ultimately this is not a desirable change. Especially tags like *_legwear which have been around since 2011. With any major change, a lot of cleanup will have to be done, and this will happen over time. We can't expect taggers to have it taken care of overnight. The beginning of the cleanup is to remove the aliases and automatically undo what we can.
As for this "right" way—people have suggested it before. It'll likely never happen. It's a restructure of the core part of the site, it's just too much of an undertaking at this point. Of course, you'd have to ask evazion to be sure.
The "right" way
Show
I think the "right" way for us to be able to handle this would be to have some sort of richer tagging system that allows tags to apply to other tags, and "belong" to other tags. Basically turning things into a full-on ontology system like WordNet or the like. That is to say I could tag an image something like "necktie{red, loose}" (just making up syntax here) with the tag "necktie" belonging to "neckwear" belonging to "accessory" belonging to "clothing" (or however detailed we wanted to go.
With that sort of system:
My "necktie{red, loose}" tag could show up in a search for "necktie{red}", "neckwear{red, loose}", "accessory{red}", "necktie", "clothing{red}" etc. allowing your search to be as general or specific as you need it.
There would be no need for umbrella or combo tags to directly apply to a post and inflate the tags there. "red_neckwear", "neckwear", "accessory", "clothing" wouldn't need to be applied to any post directly, but we'd still have the benefit of them acting like combos and umbrellas
A "bowtie{red}" tag wouldn't pollute a search for "necktie{red}".
A "necktie{red}" and a "bowtie{red}" could be tagged alongside each other and be searched for specifically (which you can't do with the umbrella system).
we could conceivably go further and assign that "necktie{red}" to a character and say search for images of "midoriya_izuku{necktie{red}}" and get posts of just him wearing a red_necktie (not others doing so).
maybe "midoriya_izuku" belongs to "male", and then we can search for "male{necktie{red}}" and get all boys wearing red neckties
I'm not sure what that might look like exactly, or how or if it could be made to be performant and user friendly (probably a nightmare), but that's been a pipe dream of mine for Danbooru since fairly early days. It'd be a super complex complete overhaul of the core of Danbooru though, and not something that might happen in the forseeable future if ever.
That all said, I'm not sure how I fall with respect to the idea of us undoing this umbrella system now. I was the opposite of NNT, where I remember being mostly opposed to the umbrella tag paradigm when we decided to go that path, but now question whether or not if it's worth the work and trade-offs to undo it.
as Shinji says, the problems faced when searching for multiple tags in the presence of images with multiple instances of characters, items, etc., is a fundamental drawback of the tagging system and affects everything, not just clothing,
using these granular attribute tags seems like a very crude solution, and i wouldn't want to sacrifice all the advantages of umbrella-tags for it.
Why not create implications to *_neckwear after removing the aliases? That way we can search for specific or unspecific colored neckwear and taggers who can't figure out what type of neckwear it is can still use the generic tag.
Just to be clear, I'm very much not 100% against this BUR, I think there pros are substantial too. I'm just ambivalent and questioning if the effort required to revert will be in our best interests or not.
Why not create implications to *_neckwear after removing the aliases? That way we can search for specific or unspecific colored neckwear and taggers who can't figure out what type of neckwear it is can still use the generic tag.
This is also a possibility, should we decide we want both the umbrella tags and the normal tags.
Shinjidude said:
Just to be clear, I'm very much not 100% against this BUR, I think there pros are substantial too. I'm just ambivalent and questioning if the effort required to revert will be in our best interests or not.
You could say this about virtually any large-scale improvements we want to make. It's kind of an annoying argument because putting things off because they're "too much effort" just means it'll be even worse to do in the future. The best time to start is now.
Thing is, most stuff under neckwears are visually very dissimilar. And certain things aren't even counted as neckwear, despite them being neckwear, see choker.
Tags that have no visual common ground shouldn't really exist.
At some point we decided that neckwear color tags weren't worth having, but we didn't have the balls to actually nuke them, instead we just merged them into color_neckwear. This is the worst of all possible worlds. If you don't like these tags, then we didn't actually get rid of them, we just renamed them into something worse. And if you do like these tags, now you have to do cumbersome searches like black neckwear+necktie to find black neckties. Which nobody does, both because it's a hassle to actually type out, and because most people never think of it. It's not even an issue of tag limits or precision, it's that it doesn't even occur to people to search for combinations like that.
I think most of the time if we want to get rid of a set of color tags, we should just flat nuke them. This soft-nuking of color tags by aliasing them away into some overly-generic umbrella tag is usually a half-assed solution that makes no one happy. The tags still exist, so the people who wanted to nuke them are unhappy, but they're too impractical to use for searching purposes, so the people who wanted to keep them are unhappy.
I edited the above BURs to only apply to solo posts for now. I noticed that there's a lot of untagged bowties/neck ribbons in necktie posts, so while those non-solo posts will still be handled automatically I'll go through them with local scripts first so that I can look at them more closely.
I'll keep these two BURs up for a day or two to make sure there's no unforeseen issues before approving them.
In the meanwhile, the old aliases have been undone, so the various color tags are populated again with the posts they had 4 years ago. See red_necktie for example.
This was suggested in the discord. These tags shouldn't exist, they mean nothing. This is a quick and easy way to move the problem somewhere else where it can be sorted out with calm.
This was suggested in the discord. These tags shouldn't exist, they mean nothing. This is a quick and easy way to move the problem somewhere else where it can be sorted out with calm.
I nuked the neckwear tag from several hundred solo images where the neckwear type was already tagged earlier, with the intention of turning the tag into something like this. I guess it's time to do the same with headwear
