What leads an internet user down a path of radicalization?
A team of journalists from The Correspondent and de Volkstrant investigated and analyzed the data.
How They Did It: Exposing Right-Wing Radicalization on YouTube
We started by investigating the rise of the alt-right on more obscure forums like 4chan and 8chan as well as on the popular chat app Discord. We were soon struck by the many references extremists made to YouTube videos, and decided to explore this platform.
The amount of right-wing extremist content available on YouTube turned out to be overwhelming: There was racism, antisemitism, antifeminism, white nationalism, and conspiracies on “cultural marxism,” “white genocide,” and “the great replacement” (the idea that white people are slowly being replaced by nonwhites through birth rates and immigration).
Around the same time, researchers began to worry that YouTube’s recommendation algorithm was exacerbating the spread of extremist content by slowly pulling viewers into a rabbit hole of hate. The recommendations that popped up when users were watching videos would slowly become more extreme: A user could start out with a left-leaning video on racism and slowly but surely end up, through a series of recommendations, watching right-wing extremist content.
In the end, we compiled a list of 1,500 channels, about evenly spread on the left-right spectrum. YouTube has a very liberal API with which you can query the database for a lot of metadata. We wrote extensive software (packaged in a reusable Python library) to examine:
- 600,000 videos
- 450,000 transcripts of those videos (by using YouTube’s automatic closed-captioning service, which is not available for all videos)
- 120 million comments on those videos
- 20 million recommendations automatically generated from viewing those videos
That’s a lot of data — about 100 GB. But what to do with it?