How does Medium “Top Stories” algorithm work?

(Notice title does not say How Medium “Top Stories” algorithm works. We’re posing a question here, people, not promising answers. OK maybe a few tiny answers.)

This post mostly written so I can tweet at Medium without having to squeeze everything in a single tweet.

Like other social platforms, Medium has a bunch ways for you to discover content.

Stuff like:

  • suggested stories at end of posts
  • interest/topic feeds
  • suggested content on the homepage/feed

This post is focused on homepage/feed suggestions. Specifically, the Top Stories feed.

While “Editors picks” feels hand-curated, by a human. Top stories is more of a hands-off algorithm. (As close as we get to “trending” on Medium.)

homepage suggested stories on Medium (web version)

But how does it work? What signals does it use? What window of time does it reflect?

Signals

We know from this Twitter thread between Nathan Bashaw + Katie Zhu (Medium employee) that part of the set of signals it uses are:

  • reader recommendations (Medium’s Like-ish social gesture)
  • time spent (by reader)

Time window

And we know from the Top stories feed heading, that it probably reflects activity over past 24 hours, because words.

TODAY. As in last 24 hours of activity. A useful clue. #sherlockholmes

Those are pretty interesting and useful notes. But this is a request for Medium to make a blog post about how their Top stories algo works.
Other social platforms write about their algos. Would be so cool for Medium Engineering to create one.

Some great examples by other social platforms talking about their algos:

Thanks for your consideration, busy Medium Engineering team.

For more being nosy about social platforms, follow @casualsheets on twitter.