Meta discussion about submitting aliases/implications

I think we need to have a discussion about when a proposed change should be an alias/implication versus a bulk change.

Something should be clear. Tag aliases and implications don't exist to create a beautiful hierarchy of data and relationships. They exist to make search easier. If a proposed change doesn't make search easier, then it doesn't serve much purpose.

There is a cost to adding a bunch of aliases and implications. Each additional one makes it harder to reason what the consequences are when another one is added. Reverts are hard and messy. Implications also add a space overhead to tag changes.

Whenever I have to review a proposal I have a hard time. A lack of discussion doesn't mean it's valid. And a lot of discussion typically means there's some ambiguity that's hard for me to suss out. So I feel I need to do something to raise the requirements to approve a change.

I've become dubious of aliases where the first tag has fewer than 100 posts. In my opinion these should just be changed with a bulk update. The alias is only useful when the from tag is being used constantly, either in tagging or searches. This is unlikely to happen with small count tags. And establishing a canonical tag early on means that people will just naturally use it.

I have conflicted feelings about implications that only exist to form arbitrary group tags. These group tags are only useful when someone searches for them. If two people use a group tag with 30k posts, then that group tag is a waste of space.

So I'm going to propose the following requirements:

1. For mass updates, the consequent tag must have a fleshed out wiki page. There are no minimum post count requirements for approval.
2. For alias and implication requests, the antecedent tag needs at least 100 posts. The consequent tag must have a fleshed out wiki page.
3. For alias and implication requests to be approved, there must be at least two other users who approve of it. If you can't muster that much interest, then the change probably doesn't have much utility.

I may implement a system to track tag searches that end up with no results to better data mine for potential aliases and implications but for now I'd like to start with this.

How do you intend to track approvals? Should the users post in the topic?

Are you sure you will be able to datamine alias usage? Since all the aliases are now automatically resolved in autocomplete, there should be nearly zero searches containing aliases, but that wouldn't mean they are not used.

Also, perhaps aliasing licensed copyright names en->jp should be considered a special exception, as it should always make the search easier for users who are only aware of translated version. Same goes for localized character names - Ace Attorney series might be the best example.

Does this include artist aliases? Do we have to mass update them from now on?

And there's the fact that mass updates don't transfer Wiki pages like aliases do, so there's more copypasting and getting the mods to delete the pages.

Type-kun said:

How do you intend to track approvals? Should the users post in the topic?

I'd like to propose a convention I've seen used in the past, where users preface their post in the topic with either "+1" or "-1" to show their approval or disapproval. Those that don't use either will be considered neutral.

tapnek said:

And there's the fact that mass updates don't transfer Wiki pages like aliases do, so there's more copypasting and getting the mods to delete the pages.

Wiki pages can be renamed, that way there doesn't need to be any copypasting or deleting going on. I've done it before when updating tags manually, which I usually do if the tag count isn't that high.

I like the +1 or -1 idea. It would be posted in the forum topic.

If moving wiki pages is the only issue then I can add support for that. Artists especially don't make much sense for aliases. They're hardly ever disputed and rarely explicitly searched for.

With the way mass updates work, I don't think implementing the added the ability to move wiki pages with each mass update can be done easily.

albert said:

2. For alias and implication requests, the antecedent tag needs at least 100 posts. The consequent tag must have a fleshed out wiki page.

Aside from artists this is how it was for years until jxh retired. Might have been 60 posts but up until toks pointed out that you stopped following it I was still arguing for this.

albert said:

Artists especially don't make much sense for aliases. They're hardly ever disputed and rarely explicitly searched for.

I don't know about others, but artist searches are the most important kind of search for me. If the artist is good, then the images are more often consistently good. You often can't really get that with a copyright, character, or general tag search. If we are going to allow artist name changes now without the alias and resultant forum proof, then it must be reiterated that there should be some way to find the new name for those unaware of the change (most important thing in my opinion is that the old name is included in Secondary Names field of the artist wiki).

A lot of people subscribe to artists. If an artist's name is going to change, and the old one will no longer work, I think they should get some kind of heads-up to edit their subscriptions.

I don't have an investment in the larger discussion here, but:

albert said:

Artists especially ... are ... rarely explicitly searched for.

I dispute this. You're the site admin, so you may have some metrics to back up that claim, and if you do, I'll shut up, but in absence of such data, I dispute the claim that artist tags rarely feature in post searches.

Upon finding an image that I like, my first reaction, every time, is to middle click the red tag and see what else this person has to offer. The artist tag is immensely important (IIRC 4 duties of an uploader is artist tag, character tags, copyright tag, and source link), and I can't imagine that importance not translating into post searches.

Serlo said:
..., is to middle click the red tag and see what else this person has to offer.

This will work whether aliases or bulk updates are used. What will be broken if bulk updates are used (like EB and fossilnix already pointed out) are saved searches, bookmarks, links from inside or outside danbooru etc. that use the old tag.

For this reason I think artist tags should still be aliased and not bulk updated (except special cases where e.g. a new artist tag has been accidentally created (e.g. twitter vs. pixiv user name) and the tag contains only one image (or a few images all uploaded very recently); in this case I think its safe to simply change the tag without creating an alias).

albert said:

2. For alias and implication requests, the antecedent tag needs at least 100 posts. The consequent tag must have a fleshed out wiki page.
3. For alias and implication requests to be approved, there must be at least two other users who approve of it. If you can't muster that much interest, then the change probably doesn't have much utility.

2. => For aliases (except artist aliases, see above) I agree. For implications I'm not so sure. I find it hard to argue that e.g. beer and sake implicate alcohol, but with the new rules an implication request for vodka to implicate alcohol would be denied because it contains too few posts. This seems awfully inconsistent and strange.

3. => Will this also be required for "obvious" implications like colored_thing => thing. I would not be surprised if nobody really bothers to open such a request to post an answer because it seems so logical that no discussion is / seems to be required.

SD-DAken said:

This will work whether aliases or bulk updates are used.

You misunderstand. I'm not arguing for or against anything to do with the main discussion of this thread. I felt that albert's comment about artist tags being "rarely ever searched for" was worth taking by itself and disputing.

SD-DAken said:

3. => Will this also be required for "obvious" implications like colored_thing => thing. I would not be surprised if nobody really bothers to open such a request to post an answer because it seems so logical that no discussion is / seems to be required.

You'd be surprised at how many people write "LE OBVIOUS LEL" only to get hit with several counterexamples and past forum topics against the alias/implication. People focus waaaay too much on the tag name and don't even do basic fucking research like looking at the tag and seeing the big picture.

Should we look at the tag counts for the antecedent and consequent for aliases and implications before weighting in on them, or is that something that albert will be doing?

Should we also be checking for Wiki pages?

I'm probably going to go with these restrictions:

• The consequent tag of an alias must have a wiki page
• Both tags of an implication must have a wiki page
• In general aliases for tags with fewer than 75 posts will not be approved. These should be submitted as bulk updates.

Thanks for the input everyone.

Wiki pages shouldn't be hard to create.

albert said:

• The consequent tag of an alias must have a wiki page

Eh, I'm gonna argue against this one. If the predicate has one it's supposed to get renamed when the alias goes through, there's code to that effect. But I agree one of them needs one.

SD-DAken said:

2. => For aliases (except artist aliases, see above) I agree. For implications I'm not so sure. I find it hard to argue that e.g. beer and sake implicate alcohol, but with the new rules an implication request for vodka to implicate alcohol would be denied because it contains too few posts. This seems awfully inconsistent and strange.

I would also agree with this, though I agree with what Hillside Moose said for number 3; +1's / backing should still be required, so any mistakes can get hit with counter examples and/or at least a lack of support.

But the 100 post rule really does create some really arbitary cases otherwise. Like SD-DAken's of beer and sake able to implicate alchohol but not vodka, when they're all alchoholic. Or, taking from a current request (topic #12595), checkered neckties being able to imply checkered but not checkered scarf or checkered hat, despite them all quite clearly being checkered items; In this case almost solely because a popular character from one of the most popular copyrights currently (Tenryuu from Kantai Collection) just so happens to wear a checkered necktie in 1/10th of her images.

I respect not wanting to mess up implications and aliases, and that making implications for less populated tags is hardly all that necessery or important... But sometimes things really can just be simple and actually obvious like those, at least enough so to be looked at. And if it's looked at and checks out, I really don't see a reason not to create an implication in such cases.