Read on Twitter
I am one of the authors of this "recently published research". The problem is not with our classifications, it's the way the labels were converted into far-right and far-left.
This paper is actually very good, but is let down by the classifications.
https://twitter.com/DavMicRot/status/1334334417125449736
This paper is actually very good, but is let down by the classifications.
https://twitter.com/DavMicRot/status/1334334417125449736
They classified channels using our old dataset (Ref 23) and ones from the Auditing Radicalization Pathways study (Ref 20).
https://arxiv.org/pdf/2011.12843.pdf
They translated labels from these studies into their categories in a really strange way.
https://arxiv.org/pdf/2011.12843.pdf
They translated labels from these studies into their categories in a really strange way.
In particular.
They coded the labels IDW/Anti-SJW -> Far Right: We consider purely anti-woke channels "right". But calling this group far-right deserves a reaction.
Conspiracy -> Far-Right: This group is too broad (e.g. 9/11 truthers, and Area 51 believers) to be far-right.
They coded the labels IDW/Anti-SJW -> Far Right: We consider purely anti-woke channels "right". But calling this group far-right deserves a reaction.
Conspiracy -> Far-Right: This group is too broad (e.g. 9/11 truthers, and Area 51 believers) to be far-right.
Social Justice -> Far Left: Pretty standard for left content. Not far-left.
Anti-Theist -> Far Left: There are two main strands of these channels that divided over Social Justice issues. Neither are far-left.
Anti-Theist -> Far Left: There are two main strands of these channels that divided over Social Justice issues. Neither are far-left.
I'm also skeptical of how useful it is to reduce the content down to the left/right dimension. My co-author writes about her thoughts on classification here https://twitter.com/AnnaZaitsev/status/1334695725855195137?s=20
People have been reacting to their favourite channels being given a bokers far-right classification, but does this impact the high level takeaways from the study?
I spent a bunch of time extracting and mapping channels from their paper to find out (why not share the data?
).
I spent a bunch of time extracting and mapping channels from their paper to find out (why not share the data?
).
There are discrepancies with the # of channels table IV and TABLE XIII, so i'm using the latter. I have charted their classifications with views stats from http://trantapareny.tube (* about 10% of the channels we couldn't match and don't have view data).
Here are the results from our dataset, with a more sensible classification I am proposing:
Use Left/Center/Right from our classification. Then apply these exceptions:
Left → Far Left: Socialist tag
Right → Far Right: White Identitarian or Partisan Right Conspiracy tags
Use Left/Center/Right from our classification. Then apply these exceptions:
Left → Far Left: Socialist tag
Right → Far Right: White Identitarian or Partisan Right Conspiracy tags
Far-left and far-right, with my proposed classification, are dramatically smaller relative to other categories compared to the papers'. We also have about 8x channel coverage coverage in our latest dataset thanks to @samuel_clark (which we have shared with the authors).
Here is a full list to review the full list of channels that have different classifications and see if this matches your understanding: https://public.tableau.com/views/Homaet_al_Classifications/ChannelDifferences?:language=en&:display_count=y&publish=yes&:origin=viz_share_link
The problems with their classification are likely to impact the headline results. Are the far-right growing? Are there far-right bubbles? The study results are hard to interpret when the classifications are widely off.
Enough criticism, now for some love. They use Nielsen data which tracks the browsing usage from a large, representative sample of americans. This allows them to get at real-world behaviour on YouTube that we have only been able to speculate about before. Pretty cool!
They can find what directed users to videos. For political videos, about 40% come from recommendations, less than we have been suspecting. I'll be referencing this in the future.
With access to the historical viewing patterns of users, they can get an indication of the stickiness of a category. They find users watching mainly "far-right" content are less likely to start watching other categories over time.
I would love to see this with the categories fixed.
They also looked at viewing sessions to see if over a session of watching videos, users were influenced to watch more videos in a particular category.
They also looked at viewing sessions to see if over a session of watching videos, users were influenced to watch more videos in a particular category.
All categories were similar, with users trending towards less political videos. If YouTube was influencing users to more right wing content, we should see the proportion of those videos increase (or at least decrease-less than other categories), but they don't.
I don't think the authors are culture warriors looking to problematize IDW/Anti-Woke creators as many are accusing. I believe that they just aren't familiar with the content they are studying, and are influenced by research that is heavily biased.
I really hope to see this study re-analysed with improved classifications, and for the researchers to share the data and code as well.
/END
/END
