Thread by @krishnagade

I was an engineer on Facebook's News Feed and this is NOT how recommender systems work.

While users can set some explicit preferences, implicit user activity on the app is the bulk of the signal that gets fed into the AI systems which control & rank the feed. /thread

https://twitter.com/DavidSacks/status/1446174602879635468

So, if you're engaging with a certain type of content of a certain set of friends - the stories from those sources get ranked higher than others. This is true with Facebook or YouTube or any other recommender system. /2

These systems take activity events from a user’s activity history as input and retrieve hundreds of potential candidate stories/videos to show to the users in their feeds. /3

These potential candidates are coming from users or content sources that are generally relevant to a given user's taste profile with high precision. /4

The similarity between users is calculated based on features such as IDs of video watches, clicks, comments, etc.

Deep Neural Networks are employed to learn high dimensional embeddings for each story, video, user, etc. /5

How these models work internally to match the content is a BlackBox because we're talking about large-scale similarity and ranking calculations in a high-dimensional space that is not interpretable. /6

This is the reason my team back then worked on tools like "Why am I seeing this?" and which also inspired me to start @fiddlerlabs.

Algorithmic governance is an important topic and civil society needs to wake up and hold large corporations accountable
krishnagade.medium.com/algorithmic-ju…