Recommended for you by YouTube: racism, antisemitism and misogyny – How YouTube fuels right-wing radicalization

2020

Recommended for you by YouTube: racism, antisemitism and misogyny – How YouTube fuels right-wing radicalization

Category: Best data-driven reporting (small and large newsrooms)

Country/area: Netherlands

Organisation: de Volkskrant, de Correspondent

Organisation size: Big

Publication date: 2 Aug 2019

Credit: Dimitri Tokmetzis, Hassan Bahara, Annieke Kranenberg, Leon de Korte, Mirjam Leunissen

Project description:

We investigated what role YouTube plays in the real-life radicalization of viewers. In a collection of written articles, data visualization, videos and a podcast, we shed light on the most important YouTube channels of the ‘reactionary right’. We uncover their connections around themes like antisemitism, anti-feminism and white supremacy and show how this can easily lead viewers down a rabbit hole of increasingly extremist content. To sketch the full picture, we did not only analyze 600.000 videos and 120 million comments from 1500 YouTube channels, but we also tracked down a handful of the commenters that expressed increasingly extremist views.

Impact reached:

Unlike other research, we did not only look at the videos that the YouTube algorithm recommends, but also unearthed the underlying community of reactionary right channels that follow and feature each other. Moreover, we reconstructed the journey of users through this network using tens of millions of their comments and interviewed some of these people about the development of their views.

Techniques/technologies used:

To uncover the network of reactionary right channels, we compiled a list of YouTube accounts that are considered to be extremist right by anti-fascism experts, academic researchers and various media sources. Using the ‘YouTube Data Tool’ of the Digital Methods Initiative we then collected all followers, subscriptions and featured channels of these accounts. We filtered the resulting collection of channels by hand and iterated this search procedure several times. Of the final 1500 channels we collected the videos, comments and other additional information from the YouTube API, using Python scripts. We also transcribed 400.000 videos using the youtube-dl Python library. Data on the monthly number of views and subscribers of the channels were obtained from Socialblade, which also provides a measure for the influence that popular channels have.

We analyzed these data with the help of statisticians, media scientists and algorithm experts, partially through two hackathon days in september and october. For our understanding, we also simply watched hundreds of the most popular videos we surfaced.

What was the hardest part of this project?

The hardest part was to reconstruct the journeys that viewers follow into the increasingly extremist corners of the YouTube universe. For this, it was essential to not only look at the videos themselves, but also the content of the comments. And to track down and interview some of the frequently anonymous commenters.