The VKontakte social network team announced the launch of a special filter to combat hate speech, which became available to community administrators.
The neural network filter can be enabled and disabled in the community settings at any time, while it is disabled by default. The function has already appeared among the owners of 50% of communities, and will soon be available to everyone.
At the moment, VKontakte’s algorithms are able to identify direct threats – wishes for death, promises to harm health and threats of reprisals.
The developers are going to continue testing the filter: in the future, it may learn to recognize other signs of intolerance, for example, bullying based on nationality or religion. Filtered posts with the ability to restore them are available to community leaders, but a test run showed that screened comments are returned only in 1% of cases.
Filter testing began in November 2020. During this period, the function has already been used by the leaders of over 10% of all VKontakte communities – this is 13 million public pages, groups and events. As the developers note, now the neural network correctly detects threats in 80% of cases.