Introducing Sofiechan Prototype Tag System

Over the past few weeks I've managed to find a few days in which to build a prototype tag system for sofiechan. In comparison to other forums, this will be our equivalent of "boards" "subreddits", etc.

Scroll down to the bottom of any index page (there is one for each tag now) to see a full list. See the top of any thread for which tags it has been added to, and for a way you can make your own tag recommendations.

The idea is that any thread can be tagged in or out of multiple tags, depending on where it is considered on topic. This means threads are "cross posted" so there will be a lot more interaction between tags than there are between boards and subreddits. As usual for sofiechan, these judgements are crowdsourced. Please help us categorize threads into the correct tags.

The UI should be intuitive, but in case it isn't, you click the little question/check/x button next to each tag to toggle through your possible judgements (unknown/ontopic/offtopic). You can add recommended tags using the +tags dropdown. Do ask questions if anything is unclear.

There's still a lot of work to do here, but I wanted to get something running in production to see how it plays. In particular, we don't yet have a way to collectively suggest and curate tags themselves. I'm just adding them manually to the database. But the plan is that there will be a way for new tags to be born, and obsolete or useless tags to die. For now, please suggest tags and changes to tags and argue about them so we can come to a good core set.

For now, the front page index is every single live thread. In future, that will be pared down to something more like the top 20 (with proper time decay and whatnot), with the overflow being indexed on the other tags. This will require some cleverness in making sure the tag index is actually covering, but that's for me to figure out. The long term idea being that the conversation is spread out among many tags, discoverable by traversing the tag landscape in various ways.

In the distant future if tags start to want to take on their own cultures and do their own expert curation I would like to have something like per-tag influence and tag leaders, but these are political questions we don't have the scale to properly discuss yet. For now, tags can at least help us define what we want to talk about, and organize our conversations.

Alternative future direction: Tags can be algorithmically managed, in a way analogous to the advisory voting on posts, where admin makes ultimate calls but user input on tags provides the data for an algorithmic determination.

>>2410
That is precisely the plan. In fact it is basically already the case. There are only a few minor things left for full integration of the tag system into the advisory voting taste machine. Basically all of you have enough signal to more or less unilaterally do things with tags unless others disagree. Tags being closer to factual judgements are way more sensitive and trusting than value judgements. So do feel free to make your judgements.

But algorithmic/advisory curation of tags as a set is also planned. I want basically anyone to be able to spawn a new tag and start adding to it, and then have it compete with the other tags for limited space in our conversational taxonomy. Tags are most useful when they are a well curated set. An ideal tag covers a topic that has some natural existence in describing our discussions, which people will be interested in following and using, which inspires good posts, and which uniquely indexes threads either missed by other tags or only hit by large tags in which they would be lost. These requirements plus direct follow/hide on tags will give us the signals to algorithmically curate a killer tag hierarchy. This is going to be a very powerful feature when fully completed.

>>2412
The tricky bit is that new tags can easily be unintentionally duplicative or overlapping, as it can be hard for a user to do a full survey of content and the existing tags.

A good tag is like a high-value vector embedding for the content of sofiechan. I was thinking that one could use vector embedding techniques to figure out when two tags are very close and should be merged, thereby reducing the number of tags and increasing the value of the remaining ones.

This merging operation could be a periodically run clean-up operation, like a semantic garbage collector. It would not compete with the voter advisory algorithm.

>>2417
Yeah basically this is what i was getting at when saying tags should index content that otherwise isn’t well indexed. The corollary is that it doesnt count towards that tag’s utility if it is duplicative. The vector embedding idea is interesting. We’ll have to try a few things to get the between-tags competition right.

For now, lets figure out which of the existing tags are the useful ones. Some are definitely duplicative or just not going to be useful but i dont know which exactly.

It might be possible to give users the 'freedom of impression' and set up their own scoring:
https://www.techpolicy.press/protecting-freedom-of-thought-by-deciding-what-information-you-consume/

>>2433
Its very important IMO that users be able to define their own “feed” and reason about it. A problem with current social media is that the feed algos are opaque, not something you can reason about and discuss. I want to solve this with the tag system first of all. Tags reify the feed categories into something human legible that can be reasoned about and discussed. “I like this tag” “i dont like the people who post in this tag” etc. The feed algorithm in that sense becomes a territory to be navigated socially and rationally rather than something that is being done to you subrationally. Rational agency is important. I think a great tag system can go a long way here.

But there is also the more vague and algorithmic concept of value that we use to rank and display things within the feed(s). Again agency is important and i think being able to relate to it collectively and socially is part of that. So for now i’m going with a collective public feed algorithm taht allows us to prioritize as a community but also to have a shared perspective that we can discuss and criticize. But maybe we will want to have more individualized predictive value (“the algorithm thinks you specifically will like this”) and ways to deliberately shape that. Im not sure yet. I want to see how the tag system plays out before i do that.

>>2438
Okay, so in your opinion, tags for posts are defined and debated upon by the community.

They could be private and modified by the readers for their convenience. Or they could be objective and defined by the poster.

One could even have all these definitions working in parallel -- the 'market' of the impression algorithms will show what works best.

The only thing to be adamant about is the first-time user experience: without that, there's no path to criticality (of impact).

>>2445
Yeah maybe when you tag something it should appear under that tag for you, with the community default following from the consensus on that.

Do you have ideas around what makes a great user experience?

>>2447
Let me wear the hat of instrumentalism:

Great user experience leads to sustained growth and engagement -- which leads to growing strength of the movement.

Indiscriminate growth and engagement in membership can actually reduce the strength of the movement.

What is the point of tags if long threads do not allow posting

>>2449
I don't see the connection. Tags are for organizing the index of threads so we can have multiple conversations going on at once. The thread length limit is so that threads don't get too long to read and navigate. The thread limit might be a bit busted right now.