Matthew Green
Does anyone who used to work at Twitter want to tell us how the algorithm worked, now that the whole thing is about to burn down anyway?
Nov 12, 2022 at 9:59pmWeb
Show threadKapitän Nemo Nov 12, 2022
@matthew_d_green can someone who used to work at twitter leak the source code? So we can create own twitter...
Show threadrx13  Nov 12, 2022
@blaubaron @matthew_d_green
Show threadChristopher PattonNov 13, 2022
@blaubaron @matthew_d_green mastodon seems purdy good
Show threadDavid ViennaNov 12, 2022
@matthew_d_green Pretty sure it involved tea leaves and chicken innards.
Show threadStephen ShanklandNov 12, 2022
@matthew_d_green I've been wondering if it's possible to apply an algorithm to Mastodon for those of us who want one to manage the mess and prioritize the good stuff. Chronological feed works best with only small number of followed accounts, I suspect.
Show threadYingtai has moved!Nov 13, 2022
@stshank @matthew_d_green I saw a conversation about the possibility of an opt-in algorithm over here recently. The OP is a new arrival from Twitter but he's also a researcher in this area.
Show threadAmit SerperNov 12, 2022
@matthew_d_green I have the pseudocode right here:
If tweet.RageInducing == true
{
PromoteTweet()
}
Show threadtaco Nov 12, 2022
@matthew_d_green
if(matthew.will_be_sad_to_see(post) show_post(post)
Show threadtaco Nov 12, 2022
@matthew_d_green btw that isn’t an implementation for you, the whole birdsite was based on what made you sad. Pretty weird but what are you gonna do
Show threadItzTrainNov 12, 2022
@matthew_d_green hahah hell yea! I want the cheat codes!!!
Show threadNemo_bis 🌈Nov 12, 2022
@matthew_d_green Isn't it all transparently published already? 🙃 https://www.usenix.org/system/files/conference/hotcloud18/hotcloud18-paper-grewal.pdf
Show threadmkbNov 12, 2022
@matthew_d_green Often with large codebases you get a blind man and the elephant situation: Many people understand pieces but nobody groks in fullness.
Show threadrrgNov 13, 2022
@matthew_d_green I don't know about Twitter but there was an interesting Gizmodo piece from May about Facebook News Feed algorithm (with a big collection of actual leaked documents which I didn't have time to read): https://gizmodo.com/facebook-news-feed-ranking-algorithms-how-they-work-1848814459
Here's What the Facebook Papers Say About the Ranking Algorithms That Control Your News Feed

Facebook employees knew that a computer-curated feed increased the time users spent on the social network—and that it led to unhealthy behaviors.

Gizmodo
Show threadolistikNov 15, 2022
@matthew_d_green couldn't they face legal repercussions? 🤔
Show threadDavide Eynard (+mala)Nov 20, 2022
@matthew_d_green well, what many tried to explain when Elon was talking about opensourcing it, there is no “the algorithm” unless you consider it as the orchestration of many different ones, trained on different features. Also, my personal opinion is that these algos (from any social network actually) are not useful “outside” their platform as they use features a regular user cannot access and their ultimate goal is some KR which is not necessarily “being good for the users”.
Show threadDavide Eynard (+mala)Nov 20, 2022
@matthew_d_green currently I am more interested in building a new one, which just works on public data and comes with plenty of docs/examples one can play with. It does not have to be perfect, just provide some easily interpretable criterion to rank posts differently from reverse chronological, and allow people to understand the basic principles it works on.
Show threadDavide Eynard (+mala)Nov 20, 2022
@matthew_d_green source: I used to work at Twitter, and tried to propose this idea while I was there