Yesterday Facebook another report on social media community standards enforcement. The company provides data and metrics for the time period from January to March, and pays special attention to the total volume of prohibited content that ends up on Facebook, as well as the percentage of it that the social network successfully removed at the stage of publication or at least before it had a chance to see it. random social network user. Facebook notes the special role of artificial intelligence (AI), without which the company simply would not have coped with filtering such an insane amount of content.
According to Facebook, artificial intelligence and machine learning have helped to significantly reduce the amount of prohibited content on the social network. In six of the nine categories tracked in the report, the company reports that it was able to pre-detect 96,8% of inappropriate posts using AI and remove them before anyone could see them (compared to 96,2% in 4th quarter of 2018). In terms of hate speech, the report says that artificial intelligence helped identify 65% of the more than four million such messages removed from Facebook each quarter, up from 24% just over a year ago and 59% in Q4 2018.
Facebook is also using AI to identify posts, personal ads, photos and videos that violate regulations on advertising and selling illegal goods such as drugs and firearms. In the first quarter of 2019, the company said it had taken action on about 900 drug-related posts, of which 000% were detected using artificial intelligence. During the same period, Facebook also detected and removed approximately 83,3 posts about the sale of firearms, of which 670% were processed before moderators or users could encounter them.
Various improvements in artificial intelligence algorithms have led to a decrease in the overall volume of prohibited content viewed on Facebook. The company estimates that for every 10 visits to the social network, only 000 to 11 users have encountered pornographic content, and only 14 may have noticed posts containing cruelty and violence. When it comes to terrorism, child nudity and sexual exploitation, the figures are even lower. Facebook reports that in Q25 1, for every 2019 views on the social network, less than three were of similar content.
“With proactive tracking of posts with violations, this technology allows our team to focus on identifying trends in how violators try to circumvent our restrictions,” Guy Rosen, Facebook vice president of content security, wrote in a blog post. "We continue to invest in technology to expand our ability to detect inappropriate content across languages and regions."
Another area in which Facebook uses artificial intelligence is spam accounts. At F8's annual developer conference in San Francisco, F700 CTO Mike Schroepfe revealed that in a single quarter, Facebook blocks over a billion spam accounts, over 1,2 million fake accounts, and tens of millions of pieces of content containing nudity and violence. According to him, AI is the main source of detection and countermeasures in these categories. In exact numbers, Facebook suspended 4 billion accounts in Q2018 2,19 and 1 billion in Q2019 XNUMX.
Source: 3dnews.ru