The role that YouTube and its behind-the-scenes recommendation algorithm plays in encouraging online radicalization has been suggested by both journalists and academics alike. This study directly quantifies these claims by examining the role that YouTube's algorithm plays in suggesting radicalized content. After categorizing nearly 800 political channels, we were able to differentiate between political schemas in order to analyze the algorithm traffic flows out and between each group. After conducting a detailed analysis of recommendations received by each channel type, we refute the popular radicalization claims. To the contrary, these data suggest that YouTube's recommendation algorithm actively discourages viewers from visiting radicalizing or extremist content. Instead, the algorithm is shown to favor mainstream media and cable news content over independent YouTube channels with slant towards left-leaning or politically neutral channels. Our study thus suggests that YouTube's recommendation algorithm fails to promote inflammatory or radicalized content, as previously claimed by several outlets.
Metadata: arXiv. Abstracts retain their original authors’ rights.
In the conversation
Original HN comments, with labels from simple text rules. A label describes the whole comment and may miss context or irony. Read the discussion to make up your own mind. About the labels
I know a few journalists really want to push this narrative, but I think the more it's been carefully studied, the more it's been debunked. E.g:
https://arxiv.org/pdf/1912.11211.pdf
"After conducting a detailed analysis of recommendations received by each channel type, we refute the popular radicalization claims. To the contrary, these data suggest that YouTubes recommendation algorithm actively discourages viewers from visiting radicalizing or extremist content. Instead, the algorithm is shown to favor mainstream media and cable news content over independent YouTube channels with slant towards left-leaning or politically neutral channels."
I'd appreciate any links you have to this evidence. I also found this:
> After conducting a detailed analysis of recommendations received by each channel type, we refute the popular radicalization claims. To the contrary, these data suggest that YouTube’s recommendation algorithm actively discourages viewers from visiting radicalizing or extremist content. Instead, the algorithm is shown to favor mainstream media and cable news content over independent YouTube channels with slant towards left-leaning or politically neutral channels. Our study thus suggests that YouTube’s recommendation algorithm fails to promote inflammatory or radicalized content, as previously claimed by several outlets.
Here's a more accessible writeup[1].
[0]: View of Algorithmic extremism: Examining YouTube's rabbit hole of radicalization https://arxiv.org/abs/1912.11211
[1]: https://mark-ledwich.medium.com/youtube-radicalization-an-au...
Not sure why this is downvoted. The YouTube radicalization pathway is fairly well documented, e.g. [1][2][3]. Even if we can't all agree on the specific mechanism of action, the phenomenon certainly exists.
[1] https://arxiv.org/abs/1912.11211
[2] https://mark-ledwich.medium.com/youtube-radicalization-an-au...
[3] https://www.wired.com/story/not-youtubes-algorithm-radicaliz...
Some of this blandness may be by design, at least when it comes to political content. YouTube caught a lot of flak in 2018 for allegedly being a "radicalization pipeline", but research[0] shows that their "next video" recommendations consistently direct users away from fringe content and towards mainstream media outlets. One of the authors of that paper made a website to visualize their results[1], which is pretty neat.
The research in question was limited by not using logged-in accounts with a significant view history, but it's still the best attempt to quantify the YouTube algorithm's influence that I've seen so far. Mozilla's initiative sounds interesting, but the fact…
Read the full comment
that they're asking users to only report negative interactions rather than getting a representative snapshot of all YouTube use makes me suspect that Mozilla is approaching this from the perspective of advocacy, not scientific inquiry.
---
[0] https://arxiv.org/pdf/1912.11211.pdf
[1] https://recfluence.net/
Doing what you suggest has an immense issue of selection bias. You go into youtube trying to see if it recommends right wing videos, and you're going to be a lot more likely to click on right wing videos. Which then informs the algorithm that you're interested in right wing videos. The correct way to test this is to write a program, not to just watch youtube while logged off.
I used selenium to do exactly this. Start at a video, and do a breadth based search based on recommendations. No such preference for far right videos exist. It's frequently get into a hole of certain topics (e.g. first recommendation is on knitting, subsequent recommendations have a higher chance of being about knitting, and so on), but there was no substantial representation of right wing content. And it's not just me with a selenium driver. Other researchers have examined youtube and reached the same conclusion: https://arxiv.org/pdf/1912.11211.pdf
My money is on it has the opposite effect, and by banning these people it moves their fans onto fringe platforms that will shuttle them towards even more bizarre/fringe/extreme views.
Here's a study that shows it pushes people towards more moderate videos.
https://arxiv.org/pdf/1912.11211.pdf
> I didn't have to look very far; Facebook did the study itself, internally, and executives directly chose to ignore it: https://www.theverge.com/2020/5/26/21270659/facebook-divisio....
While this is indeed a compelling data point, I want to point out 2 things:
1. We have no way to verify the validity of the methodology, since the study is opaque to us.
2. We have no way to verify the veracity of the results, since the study is not peer reviewed.
And while you didn't have to look very far, perhaps if you look even a little farther, you'll find that there are conflicting studies (in published peer reviewed research):
"We combine nine previously proposed measures to…
Read the full comment
construct an index of political polarization among US adults. We find that the growth in polarization in recent years is largest for the demographic groups least likely to use the internet and social media. For example, our overall index and eight of the nine individual measures show greater increases for those older than 75 than for those aged 18–39. These facts argue against the hypothesis that the internet is a primary driver of rising political polarization." [1]
"Contrary to conventional wisdom, my analysis provides evidence that social media
usage reduces mass political polarization."[2]
"After conducting a detailed analysis of recommendations received by each channel type, we refute the popular radicalization claims. To the contrary, these data suggest that YouTube's recommendation algorithm actively discourages viewers from visiting radicalizing or extremist content."[3]
So this is by no means conclusive, and very much debatable. To repeat: the most charitable argument for "being proud" to work at FB is if you believe in classical liberal principles of free speech, and don't believe that social media itself creates polarization/division, it simply reveals it.
[1] https://www.nber.org/papers/w23258
[2] http://rubenson.org/wp-content/uploads/2015/10/barbera-tpbw....
[3] https://arxiv.org/abs/1912.11211
Can you link to Tufekci's research? I can only find the NYTimes op-ed. Furthermore it's worth nothing that other studies determined the opposite, YouTube disproportionately recommends left leaning content: https://arxiv.org/pdf/1912.11211.pdf
Fast compilation is important when compilation occurs at runtime, such as query compilers in modern database systems and WebAssembly virtual machines in modern browsers.
The author presents two tricks to accelerate depth-first search algorithms for a class of combinatorial puzzle problems, such as tiling a tray by a fixed set of polyominoes.
Jon Kleinberg, Sendhil Mullainathan, Manish Raghavan
Recent discussion in the public sphere about algorithmic classification has involved tension between competing notions of what it means for a probabilistic classification to be fair to different groups.