Open source content serving algorithms? We’d need to classify the content and have a server that recommends based on the algorithm. Also, user tracking to some degree (but maybe that can be handled locally?)
I don’t see how it couldn’t theoretically be done on the user’s end with the right client. Choose a certain feed and the client takes the information shared in the incoming post data like the title, description, community, vote count, and instance. Then isolates each post, classifies it, ranks it based on user preferences stored locally like upvote and downvote history, and the client then chooses the order of posts to show to the user. And if you wanted to go there you could even have a local simple machine learning model creating descriptions of image posts to cover everything.
Granted I actually know nothing about programming, but I don’t think you’d actually be processing that much data if you kept the algorithm simple. All it actually has to do is just choose a ranking based on metrics and keywords and assigned values. It also doesn’t have to achieve maximum retention or have single digit millisecond load time, it just has to give people a customized experience.
The problem is that all this overhead and maintenance would require some form of monetization, like injecting ads into the feed. Something like that has almost no demand right now, because the options we have for sorting are good enough and the people who want custom algorithms don’t know what federation means and aren’t paying. And honestly I think it might be better without it, because personally I don’t want lemmy to go mainstream and am happy with where it is now. I sort by top of the day in All, which basically crowdsources ranking anyways.
I think a simple ‘popular’ or ‘trending’ feed wouldn’t need user tracking. Or if it was really needed to be customized maybe users could opt into share what topics they are interested in when they sign up
Open source content serving algorithms? We’d need to classify the content and have a server that recommends based on the algorithm. Also, user tracking to some degree (but maybe that can be handled locally?)
I don’t see how it couldn’t theoretically be done on the user’s end with the right client. Choose a certain feed and the client takes the information shared in the incoming post data like the title, description, community, vote count, and instance. Then isolates each post, classifies it, ranks it based on user preferences stored locally like upvote and downvote history, and the client then chooses the order of posts to show to the user. And if you wanted to go there you could even have a local simple machine learning model creating descriptions of image posts to cover everything.
Granted I actually know nothing about programming, but I don’t think you’d actually be processing that much data if you kept the algorithm simple. All it actually has to do is just choose a ranking based on metrics and keywords and assigned values. It also doesn’t have to achieve maximum retention or have single digit millisecond load time, it just has to give people a customized experience.
The problem is that all this overhead and maintenance would require some form of monetization, like injecting ads into the feed. Something like that has almost no demand right now, because the options we have for sorting are good enough and the people who want custom algorithms don’t know what federation means and aren’t paying. And honestly I think it might be better without it, because personally I don’t want lemmy to go mainstream and am happy with where it is now. I sort by top of the day in All, which basically crowdsources ranking anyways.
I think a simple ‘popular’ or ‘trending’ feed wouldn’t need user tracking. Or if it was really needed to be customized maybe users could opt into share what topics they are interested in when they sign up