Reverse-engineering based upon results.
It was a bad guess
It's mostly just a bunch of GMs and press releases from top-10 gigantic accounts and the occasional picture of someone's arse.
Could the algo be improved? Or any other way to improve the state of things?
nostr:nprofile1qyv8wumn8ghj7enfd36x2u3wdehhxarj9emkjmn99uq32amnwvaz7tmjv4kxz7fwv3sk6atn9e5k7tcprfmhxue69uhkummnw3ezuargv4ekzmt9vdshgtnfduhszxmhwden5te0w35x2en0wfjhxapwdehhxarjxyhxxmmd9uq3zamnwvaz7tmwdaehgu3wwa5kuef0qyfhwumn8ghj7ur4wfcxcetsv9njuetn9uqzp22rfmsktmgpk2rtan7zwu00zuzax5maq5dnsu5g3xxvqr2u3pd7ad6wwx didn't you have a method for normalizing the effect of follower counts?
It's already more sophisticated than that.
Then why is it still like Jack Calle Jack Calle Jack That's literally what it is, right now. 🤷♀️
Maybe that's because people here care about what these people say. I may be wrong, would you consider championing a fix for potentially broken nostr trending algos? We have history of trending on nostr.band, all data is available on our relay and API, construct a better trending list than what we had for yesterday, and I bet people will zap you like crazy. I will give 500k for algo that's consistently better.
Obviously, people here care about what those people have to say, but everyone follows them, already, so the trending list adds no information, and many people arriving probably don't care about what they have to say and see the trending list and assume this is just a shrine to a few personalities and a place to say incomprehensible texts full of naughty words.
Ok, construct an "onboarding" list for yesterday that you believe would work well for newcomers, that's a very needed thing that nobody solved yet.
I've created an onboarding relay "theforest", as I started with lists, but they were too static. nostr:nprofile1qyfhwumn8ghj7mmxve3ksctfdch8qatz9uqsuamnwvaz7tmwdaejumr0dshszrthwden5te0vfexytnfduhsqgxqvze3lc4mkzly6wfmclzq4qyy3gjm3u8q7wpvkk6fcdalw3mvkst5kamd has the idea for rings.
Perhaps we really can come up with something "better", that doesn't merely list the things everyone has already seen, but that aids discovery of interesting conversations. This is merely a statistical problem. We will think about it. 🤔
I think nostr:nprofile1qythwumn8ghj7un9d3shjtnswf5k6ctv9ehx2ap0qyfhwumn8ghj7un9d3shjtnxxaazu6t09uq35amnwvaz7tms09exzmtfvshxv6tpw34xze3wvdhk6tcqyrafsj7hmweg9ur7zmn6apajdg48hxuskujx53rhrux0ttjcqx84y2yrw28 and nostr:nprofile1qydhwumn8ghj7argv4nx7un9wd6zumn0wd68yvfwvdhk6tcprfmhxue69uhkummnw3ezuargv4ekzmt9vdshgtnfduhszythwden5te0dehhxarj9emkjmn99uq3vamnwvaz7t69v3jkutnwdaehgu3wd3skuep0qy2hwumn8ghj7un9d3shjtnyv9kh2uewd9hj7qg5waehxw309aex2mrp09skymr99ehhyee0qqsfnw64j8y3zesqlpz3qlf3lx6eutmu0cy6rluq96z0r4pa54tu5eq974x04 are already working on it. Can they be in on the bounty?
Absolutely. But we are talking global stuff here, not personalized. And for newcomers you have nothing to personalize on, they don't follow anyone yet. We may all agree personalized is better than global (and yet most apps still haven't adopted it?), but if you want or need global, then please show me algo that's much better, I will happily zap it and use it. Personalised is easy, most people don't follow junk, so input is high quality. Try any simple algo on global and you'll get garbage. Or maybe not, then you get zapped and everyone wins.
You assume people are constantly watching their feeds. I don't. I get a lot of value from the trending.
No, most people look at trending, is what I assume. Some of the clients have it really prominent. But if trending just looks like what their feed would look like anyway.... 🤷♀️ Who isn't following Jack, Gigi, Lyn, Odell, Calle, and Carla, or having their content reposted ad nauseum in their feed? The trending list seems to mostly be a condensing of our feeds, so that each post only appears once, rather than 58 times. 😂 So, that is value added.
The “what did I miss DVM” is orders of magnitude than the “globally” trending stuff. the globally trending stuff will always be skewed to accounts with more reach which is why GM posts by accounts with tons of followers are always there even though there’s far more interesting stuff from far smaller accounts. the globally trending stuff is recreating the same mechanisms of legacy media
These are different things, one is personalised, the other is not. Same algo I use for global would work well if I only fed it posts by people you follow. Annoyed that accounts with big reach have nothing better to say than GM, don't look at that tab. Algo is not the problem.
You can work with: 1. Diminishing returns: on followers counts, zaps, replies, ... 2. Looking at thread length in the replies as an indicator of real conversation beyond "GM, GM"
He says they're doing all of that, but it has little discernable effect. Probably because the deviation/spread in follower count is so humungous.
Yeah, the onboarding of most apps doesn't really help either with that humongous difference. I'm surprised looking at thread length wouldn't have an influence. That's the first crude algo I'd use to find actual conversation (like this one)
Well, length alone would strongly favor a GM post. Only measuring depth/width/nesting, would uncover interaction.
Yes! That's exactly what I mean with "length" (oops 🤦)
I was also thinking it'd be interesting to have a heat map that showed clusters of threads with active, lively interactions around the same topic, rather than only individual thread content. More threads with a structure like OP A B C A B D A C D E Instead of with "GM structure" OP A B C D E