Hacker News Reader

Comment by betwixthewires | Hacker News Reader

> Discoverability has always been a hot topic on Mastodon. Discoverability makes or breaks a platform, as there is nothing more important to retain a new user than to let them find something interesting to stay for, as soon as possible. In 3.5, we bring a new explore page which features currently popular posts, news stories that people share a lot, trending hashtags and follow recommendations. Furthermore, for the first time, we attempt to bring people content in their own language.

An algorithm to rank posts and people? Did we learn nothing from twitter?

View on HN

Honestly, an AGPL'ed, fully transparent, shareable, forkable and self hostable algorithm to do this is a whole new ball game. Where it will land us remains to be seen but compared to the current state of affairs I want to find out.

That's a very good point, I should've considered that, I've been swayed.

I think that depends what is baked in...using an algorithm to seek things one actually explicitly ask for/configs for, versus something that gets more oblique and "predictive" or makes decisions FOR you...let's give the code a look and decide after we actually scrutinize it, rather than just assume it's reimplementing the same problems again.

This is after all, NOT twitter: https://github.com/mastodon/mastodon

Yeah you're right, I think the larger problem is an algorithm that "learns" from the user, it's a positive feedback loop that produces nasty results. A simple ranking algorithm, especially one we can look at, would be interesting.

Something to consider though: even a simple ranking algorithm is going to influence social behavior on the network. I highly doubt most people will use the firehouse feeds with this option, and once you go down this road you begin talking about social impact. It's an interesting experiment, I'm not so averse to it now, but I'm wary of it still.

Another redeeming factor is that it's not personalised, allowing for it to be researched, and preventing people getting rabbitholed away from their peers.

Also it'd be great to have multiple (open) algorithms selectable, lets say:

- Friends of friends - Keywords - Popular toots...

What we learnt from Twitter, and Facebook, and (I've never worked in those two, but I've worked in other companies) is that ranking posts in general increases user satisfaction, and how often they return.

I realise there are a number of tech people who hate it, and it should definately be configurable, but companies aren't doing it to be mean, they are doing it because it works.

Their metric for "it works" is that it makes them money. This isn't Mastodon's goal.

The metric for "it works" is whether people successfully get started and continue to use the platform. For commercial companies this coincides with them making money. Even though Mastodon is a non-profit and does not and has no way of making money from having more users, it is a valuable goal in itself. Each new user decreases the network effects pull of commercial social networks and increases the usefulness of the network for everyone on it. Especially since the userbase is never static as people leave due to natural and unnatural causes, it is important to keep an eye on new user retention.

I think we both agree there are things that Mastodon should never do, even if it increases retention, like highly addictive features.

A line needs to be drawn somewhere, and it is not clear to me that Mastodon did, as a project; because it keeps pushing what I (and other early adopters) consider anti-features in the name of new user retention.

For example, trending toots would have been unthinkable four years ago while trending tags were being designed, but trending tags eventually led to that.

The OP release notes announced "As we value safety, these new features come with their own moderation tools–nothing will show up in trends unless reviewed by one of the server’s moderators first."

Just a week later, here is a pull request adding an option to allow trends without reviews: https://github.com/mastodon/mastodon/pull/17977

Their metric isn't user satisfaction, it's user engagement as a proxy for user satisfaction, so says the sales pitch. It's obvious at this point it is not a sufficient proxy. Slot machines have a high user engagement and return rate too.

We are talking about a seperate explore tab, not your regular feed, which is still (and always will be) chronologicaly ordered.

Reply on news.ycombinator.com