Using AI to fight hate speech on social media

In just one year, Facebook increased its capacity to identify and take action against hate speech by 165%. Thanks to advances in technology, the social media network is now able to catch up to 96.8% of hate speech proactively, without relying on someone to report it.

“Safety and security is an important priority for us at Facebook, and fighting hate speech is an important part of our work. We believe that people use their voice and connect more freely when they don’t feel attacked on the basis of who they are. That’s why we don’t allow hate speech on Facebook. It creates an environment of intimidation and exclusion, and in some cases may promote offline violence,” said Melissa Chin, Content Policy Manager, Hate Speech APAC, during a press briefing with Cambodian journalists.

Facebook defines hate speech as a direct attack, such as violent or dehumanizing speech, statements of inferiority, cursing, or calling for exclusion or segregation, on people, based on their protected characteristics. These protected characteristics include race, ethnicity, national origin, disability, religion, sexual orientation, sex, gender identity or serious diseases.

Facebook estimates that out of every 10,000 views of content on Facebook, just 5 to 6 of these views include hate speech.
“When we first began reporting our metrics for hate speech in late 2017, we took action on 23.6% of hate speech content before it was even reported to us by users. Since then, our detection rate has increased to 96.8%, which means that in just over 3 years, we increased the amount of hate speech we can proactively identify using technology by more than 320%, “ said Chin.

In order to create a safe environment for people from all over the world to connect and communicate, Facebook’s Community Standards outline what is and is not allowed on Facebook . Facebook’s Community Standards cover a wide range of content types including hate speech, as well as terrorism, child safety and others. Since last year, Facebook’s Community Standards have been available in Khmer language.

Facebook uses a combination of reports from the user community, reviews by its teams, and technology, to identify and review content against its Community Standards. There are more than 35,000 people at Facebook working on safety and security, including more than 15,000 content moderators.

“Our policies are constantly being updated to reflect changes in the way language is used online. We use artificial intelligence to prioritize content that needs reviewing, after considering the potential of the content to go viral, how harmful it may be and how likely certain content is to violate our policies ,” Chin said.

More information about Facebook’s approach to addressing hate speech can be found in the Community Standards Enforcement Report.