The best algorithm for a feed is very simple.
>Show my follows + their follows.
That’s it. No real person in the network follows spammers, that is the filter. If anyone starts to follow a spammer I can simply unfollow the idiot.
This immediately results in a self correcting mass of a network. Whether we see this soon or much later, this is the most elegant way and will ultimately be where we end up.
To be clear…
I follow 300 pubkeys
10^2
My follows follow 17,000 pubkeys
There are 10,000 human MAU’s
10^4
There are 1,000,000’s of pubkeys.
10^6
My follows follow is the best and most elegant filter to approximate on human DAU’s.
Instead almost all clients offer me feeds of either my 300 follows OR the 1,000,000’s of pubkey universe.
Give me a 10^4 feed! Do it in a simple way that an old lady’s cat can understand.
Yes and no. There also needs to be a way to discover new followers and new users need to be able discover established users.
Currently this can be done by hashtags and manual browsing but at some point we will need suggestions based on topics and other analytics tools to find who to follow.
Think again.
If you are brand new and you followed 2 people, then my follows followers feed would likely include 500 people.
So it’s a massive aid to discovery. Much better than what we do now.
My follows follows initially grows very quickly and then plateaus out as you reach diminishing returns.
Ugh, my follows follows (not followers).
So you say that Iris has been doing it the right way all the time? I think so too!
- Only reason for me to not use iris as my primary client is that the android app (web wrapper) is quite slow and bit laggy.
Yes “follow distance” is a good term for it.
A follow distance of 2 captures pretty much all of the active users and filters out 99% of spammers.
This is the main “lump” of the network.
I think this is a super good idea. By default it can be this. Then give the users the choice on how many steps I can choose to go for new content. I can retroactivly change from 1 step (default) to 2,3,4 and choose my optimal step.
Then like twitter take feedback from the user to selectively not want a particular type of content. nostr:npub1gcxzte5zlkncx26j68ez60fzkvtkm9e0vrwdcvsjakxf9mu9qewqlfnj5z
This is only true if most/all clients have a way to show you where the spam message came from (from which followed follow). But yeah, I like this thaught.
