Hacker News Reader

Comment by wrkccnt | Hacker News Reader

Was skimming through the report[1] and they mention “coefficient ranking” (email starting on page 49). Did some Googling and couldn’t find anything. Anyone has a brief explanation of this?

[1] https://www.parliament.uk/documents/commons-committees/cultu...

View on HN

You have a coefficient with each of your friends on Facebook, that represents how close you are with them, based on how much you interact with them on Facebook. Coefficient ranking is probably sorting your friends list by coefficient to show you the ones you're closest with first.

You have a coefficient with every object in facebooks graph, actually. Pages, friends of friends, even apps. It’s based on interactions, sure, but you should probably think about it more as a probability of future interaction, which is why they use it for sorting.

Sounds flawed. What if I interact with someone a lot because they're a family member who shares dumb boomer memes and we argue? :)

That's actually exactly how the algorithm is supposed to work. Negative emotions are far more catchy. You see dumb boomer memes constantly, you constantly argue, but you're ENGAGED. Which means you're a juicy user for ad targeting.

Sounds flawed, yes. But what makes you think it can't include whether or not the person is a family member as part of the coefficient calculation? And, what makes you think that the way you interact with your family is indicative of the way others do or don't interact with one or more of theirs?

Do you still see ads? Yep.

Great! Thanks.

It’s actually one of the most publicly documented aspects of the company, but never under the word ‘coefficient’ because the copy-writers (the linguists deciding how Facebook talks about itself on the site and in its blog posts, a surprisingly influential group) really hate the word. They manage to rephrase every interaction into a phrase, typically “how close you are with them”.

One recent example of a TechCrunch guide on News Feed [1] where “Creator” is actually coefficient or a more recent official version [2]. There was a very early “How does Facebook work?” company blog post that explained a surprisingly large amount of things. It was the only official resource for a long time. The explanations were clear, if read the euphemism “how well you know them” as (a rather obvious for data-engineers) “we batch-compute a rating on a dyadic graph at regular intervals (yes, that means weekly)”. Without that idea that all those phrases correspond to the same constant, that blog post and its successor are rather unhelpful.

Coefficient-ranking is indeed omnipresent internally, to the point employees regularly use the word in front of civilians without realising you are not meant to, almost daily. Explaining the concept and or even saying it’s called “coefficient” is fine: those blog posts are considered clearly explaining those by the PR team. You are simply advised to phrase things in a more humane way, like the blog posts do, because that helps people understand (it doesn’t: admitting a single constant number is imperfect helps a lot people understand the issues they have with the News Feed).

It’s mainly used both in front-end for suggesting interactions and as a data-engineering filter: the list of every friendship is a trillion long.

[1] https://techcrunch.com/2016/09/06/ultimate-guide-to-the-news...

[2] https://newsroom.fb.com/news/2018/05/inside-feed-news-feed-r...

Reply on news.ycombinator.com