6 ms·
How does Facebook's engine score similarity between posts, exactly?
by DaiPlusPlus 5y ago
How does Facebook's engine score similarity between posts, exactly?
- PragmaticPulp 5y agoObviously we don’t see the inner workings of algorithms, from Facebook’s feed to HN’s upvote/downvote/flag/decay system. However, my sense is that by clicking the button I see fewer posts from the person who posted it, which is often quite effective at shaping my feed the way I want it anyway. Facebook does also seem to do some basic classification of hot topics (politics in particular) such that hiding a few politics posts results in fewer politics posts even from other people, too. Likes are upvotes, so it’s also important to hit the like button on content you want to see more of, and to not hit the like button on content you don’t want to see. I know it sounds obvious, but I know people who refuse to press Like buttons because they object to the concept of Likes. Using likes and the “hide post” features to curate a feed doesn’t take much more effort than upvotes and downvotes and it’s very effective at shaping my feed, in my experience. I think too many tech people simply ignore those features and can’t understand why their feed is only showing them content that other people are engaging with (often arguments).
- smolder 5y agoPart of this is their capability of processing image posts and giving it a text description. For example Facebook will look at a jpeg of an old man with some text on it and use image recognition to actually determine a description of "photo of old man saying 'top text bottom text'". I've seen these descriptions appear in the page and they're pretty impressive accuracy-wise. Beyond that, comparing posts for similarity is likely mundane stuff like who wrote it, associated words, the type of people that engage, and how.
- smolder 5y agoFYI, you can see these descriptions by inspecting the alt text for an image posted to FB. It's there for accessibility, I assume.