Ayase/Concepts

This page is a Todo/list of desired features for a crowdsourced and AI based curation system for imageboard archives, taking cues from -Boorus(mostly Danbooru), Wikimedia, Xenforo, and other software:

The Philosophy

• AI lifted, Human Lead:

This system is for an unending goal, and we have limited human labor at our disposal, this means that we will use or develop whatever automation we deem useful to our goals, but human curation will be considered above automated curation. This should mostly end up just be a case of human curators confirming what our automated systems think is correct, but as we are curating human generated data, it should be assume we will have to fill in gaps in it's abilities or fix errors it makes

• Robust Curation:

The data we are curating should be able to be curated in as many sensible ways as possible, as the whole point of this new system is getting knowledge from the noise.

The Concepts

• "Tag Logic"

Booru-style tagging has been considered a good idea to many who have encountered it, as seen with the various attempts over the almost 2 decades since it's inception to recreate it or reimplement it. Though it has one flaw.
 In all known implementations, a tag is just a string, and nothing else. as far as these systems are concern, each string is wholly unrelated to each other and can be added and removed in any fashion. this is fine until it is not. 
 we suggest that Tags be more structured, able to modify and act with the knowledge of the existence or non existence of each 

let's take "bow" for example

you select "bow" in "manual input", and it drops down the following: "Pose?" "Weapon?" "Clothing?"

selecting pose just puts it as "bow_(pose)"(ideally, qualifiers shouldn't actually appear to the user as part of the string, so in this case i imagine "pose" might be a category)

selecting "weapon" would give another drop down to the various types of weapons, and depending on context of other questions anwsered, might suggest a particular bow for a character or series

selecting "clothing" would give a drop down asking clarification for the pattern and color, and perhaps location of the bow

• Post and Thread Curation
• Automated Tag Suggestions

"we have determine this (is/is not) an animated file"
if animated file detected: "please select the timestamp that you are currently tagging for"

"what copyright or copyrights does this media contain?"
"out of the selected copyright(s), what character(s) appear in this media file?"
"this character is known to have more than one design/form, which of the following appears in this media file?"(include an option here that more or less goes "there is not enough information to discern", likely due to the character being nude.)
"does the character have a determined gender? select from the following if so" (this adds a count to the gender tag, so 1female/1male/1other/2females, etc would be auto generated)
"what bodytype does this character have? this character is usually [bodytype](,or [bodytype2]), if they are not depicted as such, please select from the following:"
"what is the character wearing? they typically have {list of tags}, if they are wearing anything else that is not included, please go to the (miscellaneous tags/manual input) section and add them there"