Meta & Children's Online SafetyChat: we really do recommend a LOT of high-NAC content to teens!
Thread View
████ name redacted. Different bars may be the same person or different people.
quoted in an excerpt in the archive.
Friday 25 June 2021
15:20Shilpa Modyreally late to respond to this, but thank you for sharing!
15:22Shilpa ModyI think it gives a good framework for thinking about how to be more targeted in reduction. Like instead of trying to lower likely NSC content for everyone, it might be easier to make a case for lowering it for people who are seeing a ton (especially in recs, where we are choosing to show them a lot and potentially amplifying)
15:23Shilpa Modydid a quick look into this and it turns out that at the high end, we really do recommend A LOT of high-NAC content to teens! Like for the top 10% of teens, our recs in Explore are 57% high-NAC
15:24Shilpa ModyShared an image
15:28Shilpa ModyIn the strategy doc I'm writing, I'm basically saying let's show less high-NAC content but personalize/target it so we're better balancing potential harm and preferences. The three specific ideas are 1) at the top end, put a limit on how much high-NAC content we recommend to teens (e.g. something below 57%!!!), 2) for all teens, cap the % of high-NAC content we recommend to whatever % they see in Home (e.g. if someone chooses to only see 10% high-NAC content based on who they follow, let's not show them more than 10% in recs), 3) better leverage negative feedback signals, e.g. if someone gives explicit/implicit negative feedback on some high-NAC content, show them less of it.
15:28Shilpa ModyCurious to hear what you think!
22:15████Wow that is a lot of high-NAC! I wonder what this graph looks like if we diff it against the % of high-NAC content that is in their feed.
Like do these teens see a lot of high-NAC recs because they have a specific interest in those topics (e.g. if they're a budding influencer or model)?
#silent
22:25████1) makes sense from a "make the world a better place" perspective, but might not fly unless we can bring in enough research to show an integrity-type, teen-specific harm so we can get a policy decision (like with weight loss pill ads for teens). If a teen is specifically seeking that content, it's hard to say they shouldn't be recommended what they want to find 2) makes a lot of sense and hopefully shouldn't be too hard to argue if we validate that it doesn't hurt overall relevance of content for teens (i.e. teens are ok seeing recs for other kinds of content instead). Given how popular memes and entertaining content is to teens, I think there should be plenty of other topics that teens find relevant and compelling 3) I think this will be incredibly easy to get approved. There's already a bunch of research from feed recs and transparency teams that negative feedback needs to result in discernible change, or it'll harm people's perception of FB as trustworthy. The tricky bit might isolating what the negative feedback is about do they not like the person? or is it the topic? Or just something specific about the post?
I think Feed&Rec's topical feedback controls will help a lot with that though
22:30████>#silent 3 cont) Something to think about here though is what our value add is to what F&R are already testing. Presumably they will already be downranking if you don't want a specific topic in feed; are we just validating that it works for NSC, or do we do something additional?
And for implicit signals, I'm not sure if we have done any research into correlations between meaningful engagement and high-NSC/NAC, but I think that could give some initial signal on whether it's worth looking more into implicit signals
22:30████Shared an image