Recent searches
Search options
I really hope y'all understand I'm trying to improve the experience here and in so doing make this a healthier and more attractive place to be.
I didn't want to be a squeaky wheel but you know what they say...
A Twitter like "Reply Quality Filter" is not technically feasible with the technology used for Mastodon's Federation (ActivityPub). So acting out in a "squeaky" way can't produce one.
An alternative "filter" option that already exists is the option for posts to only appear in the feeds of people who follow you. Thus removing you from the completely public arena, into one you can vet and control?
@jay it could definitely be done on the client side, or perhaps the instance side.
And, picking apart some ideas I've seen, if tags were added on by an instance as the post were distributed around the fediverse, other instances could potentially benefit from the analysis done on another server.
Now, that second part probably feels like it goes too far. Another weird fork thing where stuff is going to break in translation.
But I don't believe this is impossible.
@TechConnectify If you think it is technically possible, you can make a suggestion of how it is technically possible, so that people might be able to implement it.
) @TechConnectify I don't really think that tagging is possible, because it might be lost or misunderstood or could be changed (although then again everything could be manipulated in the federation process - I think?).
However, I saw you talking about sentiment analysis, which got me wondering about open-source solutions to it. So I had a look at your profile on your instance - and it appears like you are mostly using the official Mastodon Web client?
As far as my understanding goes, it should be possible (for someone with more insight into the web app source code as me) to hide notifications if they don't fit a given criteria. Maybe the web client could run every post through something like https://darenr.github.io/afinn and only show notifications for posts with a score higher than a value set by a user.
I also ran a couple of your posts through it, and it seems like the positive ones (or at least the ones I'd consider positive/nice) get a score >=2.
Perhaps you could test this with a couple of negative and neutral ones, to collect some sample data - in case anybody ever gets around to implementing something like this.
Obviously, this would be enough if you are using multiple web clients or the mobile app, as adding this only to a web client wouldn't stop a mobile app from notifying you.
Additionally, I'm not sure if these wordlist-based sentiment analysis tools are representative of a wider audience.
Unfortunately, most of the negative posts don't federate, so I don't see them, except for some people-who-don't-understand-that-spam-is-annoying posts, which is why I haven't tested negative posts.
Also, I think that software freedom also includes the user being free to want some sort of algorithm to sort through notifications/posts/replies. And I think the absence of this might be the reason some of my friends are hesitant to joining the Fediverse. Especially since it is already harder for people to find a community. And a cluttered home-feed is not something a lot of people I know want to deal with.
@jay
Sentiment analysis is tricky, and not a well solved problem yet.
One major problem is that the 'filter' has no way to understand the difference between "negative sentiment" and someone agreeing with you that a situation is bad. The filters tend to be based on biases of their inputs, so that, for example use of the word "Gay" is negatively biased.
The "secret sauce" of Twitter's 'quality filter' is being centralised they have more meta-data to judge a comment on.
@TechConnectify