Email or username:

Password:

Forgot your password?
dansup's posts Post Back to profile
dansup

Right now Loops has a single feed in chronological order, it's not the ideal algorithm but will work for now.

When testing on my friends, they mentioned this, they are used to an algo that is tailored to them and their behaviors (durationViewed, hasLiked, hasShared, hasOpenedCommentDrawer, hasCommented, hasFollowed)

This is the most complicated aspect of Loops, no it isn't the AR filters or Camera, it's the algo and feed.

Now the key question is: feed algorithm.

Feedback + Boosts appreciated

16 March at 13:02 | Open on mastodon.social
10 comments
csdummi

@dansup I've not used TikTok or YT Shorts, so I don't quite understand what is meant by an algorithm tailored to personality of the consumer.

How does an algorithm exhibit this behavior?

16 March at 13:05 | Open on babka.social
dansup

@csdummi The algorithm uses observed behaviors like how long you watch a video, or if you opened the comment drawer, to determine the next videos in your feed

16 March at 13:07 | Open on mastodon.social
csdummi

@dansup got it. Those are the inputs, the big question of course is how then it reaches a decision on what the next post will be.

Could these data points be used to derive a certain "interest" level in different accounts, topics, content (in the later two cases, how'd they be identified?) and then find more or less of that account/topic/content type to show next?

A "good" algorithm in this sense'd probably have to classify the content in it's database a whole lot more, to be able to suggest "similar" or "different" content.

@dansup got it. Those are the inputs, the big question of course is how then it reaches a decision on what the next post will be.

Could these data points be used to derive a certain "interest" level in different accounts, topics, content (in the later two cases, how'd they be identified?) and then find more or less of that account/topic/content type to show next?

16 March at 13:16 | Open on babka.social
3dcandy

@dansup perhaps it's time to go back to simplicity - chronological feed of your friends, pure, simple and easy

16 March at 13:06 | Open on mastodon.social
raminf

@dansup It seems like the other side of the equation is what data is available on each video post?

There's the internal metadata (creator, creation time, duration, camera, filters, etc.) Then attached metadata (description, hashtags, etc). A lot of recent energy has gone into determining the content via AI (what is actually inside the video/audio channels). Classifiers can also tag for emotions/sentiment.

Then it goes one step up and connects the content attributes to user votes, moderation actions, and misinformation/reporting data, to generate some sort of score (heat, excitement, or outrage). Some of these could be pre-calced during submission stage so runtime selection is faster.

An algo would crunch all these attributes into a "this person will like this clip" score.

Could make the algo engine onion-layered and pluggable. Market of algos for different tastes. For example, one that makes people happy, angry, informed, stupider, kills time, etc.

May also want to pick @mmasnick brain.

@dansup It seems like the other side of the equation is what data is available on each video post?

There's the internal metadata (creator, creation time, duration, camera, filters, etc.) Then attached metadata (description, hashtags, etc). A lot of recent energy has gone into determining the content via AI (what is actually inside the video/audio channels). Classifiers can also tag for emotions/sentiment.

16 March at 13:43 | Open on sfba.social
cwicseolfor

@dansup Old socials before intense customization just pushed popular posts, which could be a feed you could subscribe to by election (positive consent.) Having the mode of engagement be optional - making an active selection of what I want to see (e.g. include posts liked by people you follow?) and being able to change that at any time, having access to a plain chronological feed, is what matters to me.

16 March at 13:53 | Open on urbanists.social
CubeOfCheese

@dansup I believe that any kind of engagement with comments should not be a factor in the algorithm. Sometimes I comment because I like something, but sometimes I comment to express disagreement, and that should not contribute positively to the algorithm. Since there is no way to know the sentiment of that comment interaction, comments just shouldn't be a factor.

16 March at 16:08 | Open on mstdn.social
ophiocephalic 🐍

@dansup
Suggestion to frame the chronological timeline differently - it is, in fact, the fediverse's killer app. It's the reason most people land here, to escape the inscrutable manipulation of the capitalist services. You can't go anywhere else for a feed which is entirely of your own choosing, presented in an obvious and comprehensible way. Embrace the chronological feed

16 March at 17:48 | Open on kolektiva.social
David M. Levinson

@dansup the best algorithmic sort is alphabetical, obviously.

16 March at 18:50 | Open on mastodon.social
Bram Meehan

@dansup Chronological, but don’t want to deny anyone else another option.

17 March at 0:56 | Open on ohai.social
Go Up