3 ms·
I'm no fan at all of Facebook. Don't have it, never have. Think it should be destroyed utterly. You can find me on Mastodon.[0] That said: the company sees
by dredmorbius 5y ago
I'm no fan at all of Facebook. Don't have it, never have. Think it should be destroyed utterly. You can find me on Mastodon.[0]
That said: the company sees 2--3 billion MAU,[1] and sees on the order of 5 billion pieces of content submitted per day.
The best I understand, their measure of exposure is not items but "prevelance, that is, the number of total presentations* of a particular content.[2] Long-standing empirical media evidence suggests that this follows a power curve, where the number of impressions is inverse to the number of items. So, say, 1 might see 1 million impressions, 10: 100k, 100: 10k, 1,000: 1k, etc.
This means that a service can budget and staff for either the minimum prevalence threshold before manual review, or the total number of items granted more than some maximum unreviewed threshold. Machine-assisted filtering can help. In either case, though, mistakes will happen, and at 5 billion items/day, the number of misclassifications even at very high accuracy is large:
- 1%: 50m/dy
- 0.1%: 5m/dy
- 0.01%: 500k/dy
- 0.001%: 50k/dy
... which necessitates secondary review and additional costs, as well as, of course, malicious appeals by bad-faith actors. If the filtering system is fed by user reports (flags and the like), then malicious or simply disagreement-based flags may well trigger moderation. (Crowdsourcing has its own profound limits.)
Another element is that, especially with AI-based filtering systems, what results is determination without explanation. We know that a specific item was rejected, but not why. And in all likelihood, FB and its engineers cannot determine the specific reason either.
(I've encountered this situation more often from Google, again, as I don't use FB, but the underlying mechanics of AI-based decision systems are the same between such systems.)
The upshot though is:
- Moderation is necessary.
- It's ultimately capricious and error-prone. There are initiatives and proposals for greater transparency and appeals.
- Cause-determination is ... usually ... poorly founded.
________________________________
Notes:
0. https://toot.cat/@dredmorbius https://toot.cat/@dredmorbius Also Diaspora (see below).
1. Monthly active users. https://investor.fb.com/investor-news/press-release-details/2021/Facebook-Reports-First-Quarter-2021-Results/default.aspx https://investor.fb.com/investor-news/press-release-details/...
2. See Guy Rosen, VP of Integrity for both content and prevalence references: https://nitter.kavin.rocks/guyro/status/1337493574246535168?s=20 https://nitter.kavin.rocks/guyro/status/1337493574246535168?... I've written more on the topic here: https://joindiaspora.com/posts/f3617c90793101396840002590d8e506 https://joindiaspora.com/posts/f3617c90793101396840002590d8e...