Lemmy

14061 readers

12 users here now

Everything about Lemmy; bugs, gripes, praises, and advocacy.

For discussion about the lemmy.ml instance, go to !meta@lemmy.ml.

founded 5 years ago

MODERATORS

nutomic@lemmy.ml

[proposal] a better content-recommendation system (discuss.tchncs.de)

submitted 3 days ago* (last edited 3 days ago) by gandalf_der_12te@discuss.tchncs.de to c/lemmy@lemmy.ml

9 comments fedilink hide all child comments

I know that many people don't like complex or intransparent recommendation algorithms.

Currently, there are "subscribed", "local" and "all" categories (at least in the default lemmy web UI).

I would like to change this to include custom topics ("listings"). They are a custom way of choosing content (in case of Lemmy listings). In Lemmy, custom listings appear just like standard listings (API-wise), just that instead of "https://discuss.tchncs.de/?dataType=Post&listingType=Subscribed&sort=Active" you have "https: //discuss.tchncs.de/?dataType=Post&listingType=list:AAAA4865698@lemmy.world&sort=Active" or something.

Listings could either be lists of communities and other listings. Consider this simple text file to describe a listing for a memes-topic: (that contains 2 communities, everything on 1 domain, and another sub-listing)

c:memes@lemmy.world
c:memes@lemmy.ml
domain:memes.net
list:AAAA4865698@lemmy.world

Or they could take their data from an RSS stream or similar external source.

top 9 comments

sorted by: hot top controversial new old

[–] dessalines@lemmy.ml 8 points 3 days ago (1 children)

Multi-communities are already completed in the back-end, and front end will be done shortly.

[–] gandalf_der_12te@discuss.tchncs.de 1 points 3 days ago* (last edited 3 days ago) (1 children)

thanks! are multi-communities anything like sql virtual tables? i.e they appear as one community to the reader, but are the result of querying multiple communities

[–] dessalines@lemmy.ml 3 points 3 days ago* (last edited 3 days ago) (1 children)

No they don't use sql views, they're built using an in query (which we still have to see how performant that'll be, and is pretty much the same problem as subscribed communities, needing to sort and filter based off joined data)

You can see them in action here: https://github.com/LemmyNet/lemmy-ui/pull/3521

[–] gandalf_der_12te@discuss.tchncs.de 2 points 3 days ago

thanks!

[–] TechnoCat@piefed.social 6 points 3 days ago* (last edited 3 days ago) (1 children)

is it this feature? "Support for grouping communities / multi-communities #818" https://github.com/LemmyNet/lemmy/issues/818

[–] gandalf_der_12te@discuss.tchncs.de 1 points 3 days ago

i guess it's very similar :D

[–] ThorrJo@lemmy.sdf.org 3 points 3 days ago (1 children)

How about open source pluggable algos which can be optionally marked shareable between users and selected from a list sortable by number of "installs"

So like a user could define an algorithm to select posts, and then mark it shared, and other users could see it in a directory and try it out themselves, and optionally clone and hack on it, release a new version, etc.

As far as how the post selection algos could be defined, I'm thinking of something similar to the boolean query syntax in the Quod Libet music player, but one could also implement a more code-like syntax.

(I'm aware this is a huge ask, but I've had this idea for 8 years since my first Mastodon account, and have been too busy being oppressed by life to do anything about it.)

[–] gandalf_der_12te@discuss.tchncs.de 2 points 3 days ago* (last edited 3 days ago)

Yeah I'm personally tending to always separate data and processing.

Data means you have an underlying data source (in lemmy's case the set of all lemmy servers hosting lemmy communities), and a separate system that does search/recommendation based on the data.

The search/recommendation system would not store any data itself, but only filter and sort and process the data from the underlying servers and communities.

[–] Skavau@piefed.social 1 points 3 days ago

I already made a massive !forumverse@piefed.social for this. You will have to be on a piefed instance to access it.