"Find an Instance", search algorithms and metadata

I had a conversation over in this reply chain a couple days ago which I think presents some interesting points regarding how difficult it currently is for someone to find the “right” instance - or, from a less idealistic angle, to find information about the available instances at all.

To summarize the points made:

  • Creating an algorithm to scrape About pages for keywords would be a serious pain, especially when trying to account for all the languages.
  • There is not a lot of incentive for a single person or small group to put together and maintain a useful, searchable instance listing site.
  • Adding some kind of instance-specific customizable metadata to the API would spread out the work and the maintenance, plus better fit the fundamental underlying ideology of decentralization and customization.

Is adding some sort of topic and moderation keyword metadata feasible? Or has there been any discussion on the topic?

1 Like

I’d start by talking to whomever’s maintaining instances.mastodon.xyz - I’m not sure who it is, but they might be able to and amenable to pulling in /about/more and creating word clouds, if not full keyword search.

1 Like

that would be thekinrar, admin of mastodon.xyz

I kinda dream of a matcher kinda like OKCupid uses. You answer questions, pick what’s important to you, and based on the answers provided either by those running the instances or the communities there you can see instances that meet your needs etc.

3 Likes

I often see people stating how difficult it is to find other people interested in topics a, b, c… I always suggest to use hashtags, look at who people follow and are followed by, read profiles, etc, but it can be more tedious than a lot of people want to invest into it.

I also feel like the more popular instances haven’t really gelled around any one subject or theme. Even when an instance does say what it’s about, the collective tooting is more or less across the board, and you’ll have members that have various interests, more than what the scope of the instance might be.

So, I wonder if the effort to guide people the right direction is better focused on helping them find other people with similar interests rather than help them find a given instance, which is kind of a crapshoot at the moment anyway. Both would be ideal, but in the order to appease immediate needs, I’d think people connections probably need to come before local gardens.

This especially makes sense if the potential for autonomous account deletion comes in the future, because then picking “the right” instance isn’t such a big deal.

What kind of external resource or UI feature can help people find other people of similar interests? That would be useful thinking.

I think that the third option @InspectorCal proposed is probably the best. This should be an opt-in protocol that could have alternate implementations IMO. Lets the instance sort of decide how they want to be represented in summary. Maybe a call to action field for a single line teaser, then a short one paragraph write up to supliment the long form about page?

@butternuts I think tags may not be incredibly informative without context…
What if we gave a semi-interactive render of the current conversation around that tag on the instance for topic discoverability?
I think we could define current conversation as one with a root toot within 24 hours that has the most number of public viewable replies from that instance. We could either get creative with how we render it or just present the standard Mastodon timeline window. We’d obviously need to see this in action to see if it actually ends up giving a good ‘feel’ of the instance, but I think it could help.

@beatrix-bitrot I’m concerned that we’d end up encoding our own bias into that sort of a quiz. I don’t really know how we’d be able to make a quiz complex enough to capture all the values of the niche instances might value. Can you think of a way to do this without forcing square instances to fit their best into the roundish holes we leave in the questions?

The Mastodon instances form allows instance owners to write a short description of their instance but it is currently only displayed for the instances that are featured on the instances.xyz main page - not the list. It would be helpful if the same form allowed for tags as well so that, for example, someone could search ‘art’ and be presented with a list of art focused instances.

I don’t propose the creation of a quiz per se, but more of a system for allowing someone to answer a (potentially endless) stream of questions that are not generated by any central authority, but rather the communities/instances that make up the fediverse itself. You answer what you feel like and see what you match up with as you go along.

1 Like

I think this is an interesting concept, but at the same time that potentially endless point is a real issue. With a sufficiently large quiz, we’re going to see something looking like good sorting but it’s a brute force approach. I think we need to think about keeping such a thing manageable. We could maybe make quizzes generated out of small sets of randomly selected instances. This sort of a model would be good for casual engagement “I don’t really know what I’m looking for” user so I actually don’t think having a smaller pool of randomly selected instances would terribly affect its utility. Additionally making it create the quiz from a small pool will eliminate our need to introduce complex predictive analysis. We could even likely create a mode that let a quiz continue for a few rounds…

Example quiz algorithm:

  • Each instance provides 15 questions for which an affirmative question is considered selective and a short self-discription (perhaps the call to action + paragraph from my proposal above. Eh, EH?! :3)

  • Initial Population: uses 5 random questions from 5 randomly chosen instances (arbitrarily small) resulting in a quiz of 25 questions.

  • Selection: After a quiz comprising of 5 previously unused questions from each instance, a suggestion is given from simple tally (Yes - No, Maybe/No Comment discarded with ties resulting in a random instance from the pool of winners) with a self-description page and a “try again” button.

  • Try again mode: takes the second-pick (or top ties in the case of a tie) of previous round, marks the previous questions used and selects random instances until it builds a new set of 5 instances.

  • Repeat from Selection

However, I do think that we should treat this as essentially the same as the ‘random instances’ mode of the current instance picker and provide a ‘better’ list interface alongside it for people who know what they want.

Oh!! I like this idea, although as a separate or additional thing to the metadata thing I was suggesting.

What I’m envisioning right now after reading that is something like that “which cat is cuter” or “what’s your favorite pokemon” ranking quiz, right. Instance admins submit criteria that they think are important to their instance - OR, if you have the metadata API implemented, it could just be those things in the list, linked to the instance.

Then the person goes to the “quiz” area and it’s a screen with “Which is more important/interesting?” and two side-by-side boxes with one of those criteria in each side. The person clicks the one that’s more important or interesting, then does this for as long as they feel like or until it reaches some threshhold for how sorted the results are. Either way, there could be a link to “Show my suggestions!” or something that when you click it, shows a list of the 25 instances that are ranked the highest according to that user’s input on ranking the criteria.

If the person doesn’t like the list, they can click the link to go back to rank more of the criteria!

1 Like

There’s a “description” field provided at https://mastodon.social/api/v1/instance.json

Mostly left empty, but would be very useful to me if admins filled it. Currently I have to scrape about/more and parse it. It’s difficult to presume that admin-filled data will be there, will be useful, and will be (kind of) accurate. Metadata helps in a big way, but I think admins haven’t recognized yet the effects of having good metadata because there’s no good list out there yet that uses it.

2 Likes

I didn’t even know about this!! Thanks for the heads up!

1 Like

3 posts were split to a new topic: @PeanutGallery (outside way to get a feel for the fediverse)

It’s slow going, but I’m chomping up hashtags in preparation to build a tag cloud per instance. I think this sort of analysis (though not exactly this) is ideal to put in the server software as a cron-ish job if admins want to.

Without much setup (minimum: a checkbox), the instance would query its database, do some calculations, then simply deliver tag clouds/liked image URLs/what have you through “api/v1/metadata” or some such .

It’d have to be checked for privacy concerns, though.

I have some concerns that hashtag based metadata is going to already require a degree of cultural literacy to be useful. For instance, the #vore hashtag has become a common joke but without that supporting context it could seriously affect interest in an instance. I’m strongly in favor of self-description rather than relying on (or rather featuring in prominence) algorithmically generated tag clouds.

At least for the context i’m assuming this is going to be used for. E.G. helping a first time user choose an instance to take root in. I think this could be a useful tool for profiling communities for a user who is already familiar with the fediverse though.

The problem you note is real, but could be solved pretty easily by showing the admin what tags are currently being pulled for this purpose, and simply letting them blacklist tags they might not want representing the instance. I’d include #nsfw in the blacklist by default since it’s included automatically in every CW’d toot so it tends to rank higher than it should.

This could help.
But I still think it would be better to present tags in the context of how they are being used in discussion. Refer back to my comments “Find an Instance”, search algorithms and metadata for specific implimentation ideas, but I do think that in adding the conversational context that this could be a good ‘at a glance’ charactorization. Unless, you know, the discussion has devolved into meta jokes about meta jokes. Which happens.

We should definitely still keep the short “about this instance” description visable :3

https://instances.octodon.social has a quiz to help you find the instance for you, but it’s only one question (about how strongly-moderated you’d like your instance).

Hi. I’m the maintainer of instances.mastodon.xyz and I’m currently about to release an instance selection wizard by the end of the week.

While creating the wizard, I faced several problems because I wanted something that was for everybody and I had to remove steps several times.

Now, I just have these three questions:

  • Spoken languages
  • Instance size
  • Moderation (list of prohibited content)

Theme / tag search is now appart, with a search box on top of the list. It’s based on the theme instances admins give me at instances.mastodon.xyz/admin and on the short and full descriptions on the same page.

If there are still people looking at this thread, what’s your view on this and what improvements should I make?

3 Likes

‘More Info’ button on the list view please! :slight_smile:
I love that interface element on the random selection, but I want to be able to view that blurb in the more comprehensive view as well.

Yep that’s great idea but it is not related to this topic.

And actually as this part of the website is going to be rewritten too… That’s not really important right now.

1 Like