Hi,
If you’re like me, your probably seeing a lot of stuff you’ve already seen in jerboa
On Reddit this didn’t happen because the site takes into account how many times a post was printed and the more you’ve seen it, the quicker it would disappear from your version of the front page.
Now of course jerboa could and should do this, But I think there’s two opportunities to make this better than Reddit. On one part, putting the squarely in control of the content discovery algorithm, next, solicit user input and ask him to lend a hand in the social sorting algorithm that is voting.
So, a user voting sounds be a way to tell jerboa that “I’ve seen this” and it shouldn’t show it anymore on my feed. To prevent bias, the neutral vote should be added.
Next is giving the user more explicit control of the algorithm. When you vote up or down, you’re sorting for the community but also for yourself. Jerboa should take into account user’s voting pattern and recommend current based on what the user likes.
These voting patterns should be publicly exchanged in “out of band” communication. Jerboa could then use these voting patterns to further help with content discovery in the following way.
“My user likes X,Y,Z, after consulting public voting patterns, we can see that most users who like X,Y,Z often also like A,B,C and dislike I,J,K”
This is how Netflix, YouTube and other algorithms find stuff you like.
The difference is now, this runs on your computer. You can see your algorithm weights and edit them. Place extra filters on them and most important, swap , export, import algorithm sorting weights and exchange them with others users, craft them for specific usage and etc.
Plus of course, basic function like chronological view that doesn’t cheat or insert ads.
Algorithmic content discovery under user control is going to be the biggest user benefit of switching to Lemmy versus a private commercial centralized platform. Our data will finally serve us !
A sorting algorithm seems out of scope for Jerboa, which is just a lightweight client. The current algorithms, such as they are, are handled by Lemmy itself, and I think it makes more sense to keep it that way so that features aren’t unique to this one app. Plus that way you aren’t downloading the entire list of content and sorting it locally, which seems somewhat expensive versus just sorting it on the server where it already is.
Well, the user’s preference for sorting algorithm have to be communicated somehow from the jerboa client to the Lemmy server.
But from a user interface point of view, if the sorting is happening in the server or the client doesn’t make they much difference.
However, from a user to control point of view, I think it should be always possible for the client to do its own sorting.
There is no danger of losing chronological ordering , if the chronological sort is happening inside the user’s device.
Also there are privacy implications, I might want to train my algorithm to sort content recommendation based on what I like. That doesn’t mean I want to disclose this information to the server. In particular if I’m not on an instance that I trust not to exploit this information to leverage advertising against me.
Also, content discovery does not work well if users don’t share their preferences to each other. But the server doesn’t need to know this information. User might exchange this information ins peer to peer manner and keep the server out of the loop.
This would disempower the instance owners from the ability to tip the scales of content curation, since they don’t know what posts users like and client request unbiased sequential listing of content which the client itself will sort later.
Frankly, I prefer not to have content similar to what I upvote. I’m not here to stay in my own echo chamber, but to see different opinions, new things.
Then use chronological ordering if you want random new things.
In fact, if what you really want is “stuff not like I’ve seen before” then dislike what you’ve seen and that will train the algorithm to show you things not you’ve seen.
One way or the other, there will be an algorithm that sorts the content your seeing because your can’t read millions of posts and comments per second, your device has to decide what to show.
You choices are, you control the algorithm in your computer or someone else controls the algorithm while you use someone else’s computer (that’s the cloud).
Relay for Reddit did something I really loved. They had a hide all read or seen posts button that simply removed them from your current list. Didn’t affect visibility if you refreshed the list, posts would reappear.
Would this work on mobile as well? Does it require clicking into a thread to consider it viewed?
Jerboa should take into account user’s voting pattern and recommend current based on what the user likes.
I would absolutely not want this. I can already curate my feed exactly how I want it with subscriptions, but when I click on All I want it to be the exact same view that everyone else gets.
On the topic of hiding seen posts, yes I think this is a pretty common suggestion/complaint of Lemmy at the moment. Part of it is that there is currently barely a fraction of activity vs. Reddit or other websites. I am curious to see how the feed changes as more people contribute.
Also related, the staleness of ‘Hot’ and ‘Active’ is actually a bug and instances can temporary fix it by restarting.
You did not fully read my post. I explicitly mention a chronological view.
Content sorting algorithm are inevitable. How will you personally sort 5 million post per day ?
What I’m saying is they must be under individual user control. And taking part in content sorting is a duty of every Lemmy user.
I did read your entire post. I was referring to the general idea of an algoritm based on my voting patterns as you detail here:
“My user likes X,Y,Z, after consulting public voting patterns, we can see that most users who like X,Y,Z often also like A,B,C and dislike I,J,K”
This is how Netflix, YouTube and other algorithms find stuff you like.
I have personally never been satisfied with any popular algoritm, Netflix and Youtube are specifically awful at suggestion stuff I like.
How will you personally sort 5 million post per day ?
The same way it’s handled right now, a ‘hot’ or ‘active’ sort which is objectively based on user activity.
Maybe the sorting algorithms on other websites work for you and that’s okay, I was just voicing my opinion.
I don’t know what do say, I’ve had great experience with content discovery algorithm. But they need a lot of work in your part to be good. Your line and dislike action tune the algorithm.
If you don’t like/dislike things, it will but learn and you will get the common denominator mush content.
Going by plain numeric user activity is not going to yield good results. It is a positive feedback loop. This is how you get “popular stuff that is popular because it is popular”. It is the recipe to create Kim Kardashian !
The more content there is, the more sophisticated the content algorithm have to be. Or else you won’t find the stuff you like.
What we need is all the data, searchable in every way possible.
Lemmy already support showing/not showing seen posts.
All we need is a quick way to toggle it on or off.
I didn’t realize this! Do you know if that setting can be changed in Jerboa?
hamburger>settings>username’s settings>show read posts
Amazing, thank you!
I don’t think I would wqwnt curated content, but I wouldn’t mind something suggesting new magazines to me based on like users. The only thing I found tough on R was it was usually word of mouth to find a new group. I would kinda like recommendations based on my magazine sub’s, and the sub’s of folks who sub to my groups. Just a thought!
I don’t love the idea of a neutral vote, because it would necessitate you voting on each and every post for them to stop showing up on your feed, which is actually how the “hide read posts” seems to work right now, but without the neutral vote.
And fuck it, it may be a hot take here, but I don’t dislike the idea of suggested posts showing up on your feed. It’s a great way to find new communities you wouldn’t have found otherwise, even if Reddit was a bit… heavy-handed in its approach
But yeah other than that your idea for showing the user the actual weights used for their suggestions algorithm does sound interesting, but I’m not sure how plausible it is (assuming many of these algorithms use machine learning and the weights are basically meaningless to humans)
Yes, neutral vote is functionally the same as “hide this” but as an explicit vote that means “I have no feeling about this one way or the other”. I think a neutral vote gives useful information for content discovery. Imagine a content with 500 positive vote 25 negative vote and 25000 neutral vote.
The most important aspect of content discovery is the relation of your vote relative to all your other vote. This is how the algorithm can find things you might like based on what other things you like.
The current algorithms which lack a neutral vote, have no concept of mediocrity. It’s good or bad, they don’t understand “this is ok but meh”
I would greatly appreciate any kind of feature to handle read posts, even if it’s entirely manual (for now).
That said, I don’t have any strong opinions on how this should be implemented. I just wanted to say that the very fact we can have this kind of conversation out in the open is so exciting. Maybe in the very early days of Reddit or Twitter that kind of user feedback was appreciated, but the last decade of monolithic tech giants has been so depressing.
The last week or two has honestly been my most refreshing and invigorating time on the web that I can remember since I got my first Gmail account back when it first went open beta.
I concur, this feels like early Reddit. I really care that it don’t suck. Tired of getting evicted every 10 years by techbro dorks looking to make a quick buck
RIF had that feature. Anything I upvoted or downvoted stop showing next time I refreshed.
It’s so close to being what I need.
Right now you can hide read posts.
I just need it to hide on scroll past and a quick way to toggle read.
Also not hide my own posts at least under profile when I have hide read on.
Then it would be perfect for how I’d use it.
The problem with hide past by scroll is, every post you see, you have to interact with it or what it disappears forever.
I find that tedious and distracting sometimes I might not reject something right now, but maybe I don’t want to interact it right now.
Maybe if I scroll past two or three times without interaction, then yes, it shouldn’t show it again. Maybe even have a display of how many scroll past are left before it disappears forever. If I see a zero, and still feel I might want to see it some day, then I’ll take action.
This is exactly what I’m waiting for, Joey spoiled me over the years, so smooth at removing posts I’d already scrolled past
God joey is the best damn app ever. There hasn’t really been much talk from the dev apparently, but hopefully someone comes up with an elegant solution.
