admin said in #3313 3w ago:
Hey anons. We've got the tag word frequency stuff running smooth now in the lab, but we need to add tag renaming to really make it useful. Let's talk about how to do that.
We have a universe of bad tags right now. About half of them are not convincing topical collections. When we turn up the strength on the word frequency autotagger, those ones are going to drift even further from their names. So we need to be able to rename and de-name tags.
Renaming in the sofie way seems obvious: there will be some list of candidates to which you can add, you vote for your favorites and against your least favorites in the usual way, and we pick the best. We might add some default votes for the most significant words in the tag (see pic related. They're pretty good). The main question is one of design.
Ideally we keep it all on one page, like the tag votes on threads. You mouse over the title and the candidates appear with votes. Or maybe we have a little "word cloud" quoted list of the top words like in the image, and you can vote there or add to that. Or maybe the word cloud is there but it's distinct from name candidates, which are closer to the existing page title. Maybe only some of the candidates appear, sampled probabilistically. I'd like the top words to appear or be used somehow because they seem to carry a lot of signal. We're experimenting. Do you guys have ideas?
De-naming is another feature we're thinking about. The tag machine has unnamed ghost tags flying around in the background for various reasons, and it seems like a cool mechanism to lean in to. The ghost tags are those the autotagger is thinking about and experimenting with, but we don't want to actually have to see or have tied down by our votes. That's how new tags will be automatically discovered. On the flipside, sometimes we want to kill a tag, or the system should kill a tag on our behalf. It will become a de-named ghost. (/gnon/ /theology/ /metaphysics/ and /philosophy/ are going to fight to the death until one emerges as winner).
The de-naming might be initiated automatically when a tag is running low on content or usage or name candidates, or we might have an explicit option somewhere to vote for de-naming. If a tag gets de-named, probably all its votes die too, and we banish it back to the ghost realm to wander aimlessly in search of a new body.
When a ghost tag starts picking up on a strong unrepresented cluster of topics and there seems to be something stable there, we can automatically name it from its characteristic words and pop it back into regular existence. Maybe the ghosts that are close to incarnating will be visible somehow and you can grab one and name it and take it where ever you want.
The overall idea is a dialog between the autotagger, which does most of the work automatically, our background preferences as a community, and the more active work of people with particular ideas. The outcome should be a vibrant set of tags that catalog what we actually want to and like to talk about, and the actual subcultural distinctions in our larger community, and get recycled or repurposed if they aren't cutting reality at the joints.
Before I really put a lot of effort on this, I want to get feedback on the ideas and see what you all think about the various parts of this plan. How should we handle it?
We have a universe of bad tags right now. About half of them are not convincing topical collections. When we turn up the strength on the word frequency autotagger, those ones are going to drift even further from their names. So we need to be able to rename and de-name tags.
Renaming in the sofie way seems obvious: there will be some list of candidates to which you can add, you vote for your favorites and against your least favorites in the usual way, and we pick the best. We might add some default votes for the most significant words in the tag (see pic related. They're pretty good). The main question is one of design.
Ideally we keep it all on one page, like the tag votes on threads. You mouse over the title and the candidates appear with votes. Or maybe we have a little "word cloud" quoted list of the top words like in the image, and you can vote there or add to that. Or maybe the word cloud is there but it's distinct from name candidates, which are closer to the existing page title. Maybe only some of the candidates appear, sampled probabilistically. I'd like the top words to appear or be used somehow because they seem to carry a lot of signal. We're experimenting. Do you guys have ideas?
De-naming is another feature we're thinking about. The tag machine has unnamed ghost tags flying around in the background for various reasons, and it seems like a cool mechanism to lean in to. The ghost tags are those the autotagger is thinking about and experimenting with, but we don't want to actually have to see or have tied down by our votes. That's how new tags will be automatically discovered. On the flipside, sometimes we want to kill a tag, or the system should kill a tag on our behalf. It will become a de-named ghost. (/gnon/ /theology/ /metaphysics/ and /philosophy/ are going to fight to the death until one emerges as winner).
The de-naming might be initiated automatically when a tag is running low on content or usage or name candidates, or we might have an explicit option somewhere to vote for de-naming. If a tag gets de-named, probably all its votes die too, and we banish it back to the ghost realm to wander aimlessly in search of a new body.
When a ghost tag starts picking up on a strong unrepresented cluster of topics and there seems to be something stable there, we can automatically name it from its characteristic words and pop it back into regular existence. Maybe the ghosts that are close to incarnating will be visible somehow and you can grab one and name it and take it where ever you want.
The overall idea is a dialog between the autotagger, which does most of the work automatically, our background preferences as a community, and the more active work of people with particular ideas. The outcome should be a vibrant set of tags that catalog what we actually want to and like to talk about, and the actual subcultural distinctions in our larger community, and get recycled or repurposed if they aren't cutting reality at the joints.
Before I really put a lot of effort on this, I want to get feedback on the ideas and see what you all think about the various parts of this plan. How should we handle it?
referenced by: >>3315
Hey anons. We've got