Is twitter's algo an exercise in information retrieval or reinforcement learning? The decomposition matters. The social graph implicitly & powerfully filters relevant instances. So, I bet the algo is mostly applied reinforcement learning with likes / rts integrated in the reward.
-
Show this thread
-
"But, it's obviously information retrieval because likes and retweets signal relevance." But, information about what? I think the answer is: the current state of the world. You go to Google first for general information; you go to Twitter first for "where are we, now?"
1 reply 0 retweets 0 likesShow this thread -
But, if it was pure information retrieval, timeline shaping doesn't make sense. Anything that reorders or reweights based on your prior activity with the goal of increasing engagements is (almost certainly) going to bias information retrieval.
1 reply 0 retweets 0 likesShow this thread -
Note: This bias doesn't correct institutional / media bias. You could do that with who you choose follow. Instead, this bias skews presentation in favor of maximum salience. On a social network. Where the most obvious cues are social.
1 reply 0 retweets 1 likeShow this thread
So, twitter's algo presents the ideas of your friends and hate follows; encoded in 280 char expressions; with attached information-dense social cues; conditioned on things it expects will arouse you; with minimal protections against abusers. Does that seem like a good idea?
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.