The DOI matches its Crossref record; 22 of the 23 references it cites were found in a registry or an index; none of the 13 sources its claims cite is checked for support yet.
Auditing radicalization pathways on YouTube
Open accessAbstract
Non-profits, as well as the media, have hypothesized the existence of a radicalization pipeline on YouTube, claiming that users systematically progress towards more extreme content on the platform. Yet, there is to date no substantial quantitative evidence of this alleged pipeline. To close this gap, we conduct a large-scale audit of user radicalization on YouTube. We analyze 330,925 videos posted on 349 channels, which we broadly classified into four types: Media, the Alt-lite, the Intellectual Dark Web (I.D.W.), and the Alt-right. According to the aforementioned radicalization hypothesis, channels in the I.D.W. and the Alt-lite serve as gateways to fringe far-right ideology, here represented by Alt-right channels. Processing 72M+ comments, we show that the three channel types indeed increasingly share the same user base; that users consistently migrate from milder to more extreme content; and that a large percentage of users who consume Alt-right content now consumed Alt-lite and I.D.W. content in the past. We also probe YouTube's recommendation algorithm, looking at more than 2M video and channel recommendations between May/July 2019. We find that Alt-lite content is easily reachable from I.D.W. channels, while Alt-right videos are reachable only through channel recommendations. Overall, we paint a comprehensive picture of user radicalization on YouTube.
Overview
AI snapshot
Written from the paper’s full text.
TL;DR
This paper audits whether YouTube users radicalize across I.D.W., Alt-lite, and Alt-right channels, using 330,925 videos from 349 channels and 72M+ comments. It finds that the three communities increasingly share commenters, that users migrate from milder to more extreme content, and that Alt-right channels are reachable through YouTube’s channel recommendations even without personalization.
Key findings
- The commenting user bases of the I.D.W., Alt-lite, and Alt-right increasingly overlap: in 2018, the Jaccard similarity between Alt-lite and I.D.W. reached almost 30%, and the Overlap Coefficient of Alt-right with Alt-lite and I.D.W. reached around 50%, meaning about half of users who commented on Alt-right channels also commented in the other communities.
- Users systematically migrate toward more extreme content: among users who in 2006–2012 commented only on I.D.W. or Alt-lite content (227,945 users), by 2018 around 10% were lightly exposed to Alt-right content and roughly 4% were mildly or severely exposed, about 9K users total; among those who in 2017 commented only on Alt-lite or I.D.W. videos (1,251,674 users), approximately 12% were exposed, more than 60K users.
- In 2018, roughly 40% of Alt-right commenting users could be traced back to cohorts that had commented only on Alt-lite or I.D.W. videos in the past; media-channel users never surpassed 6% for any exposure level. For light exposure in 2018, 23.3% traced back to prior Alt-lite-only commenters versus 7.6% to prior I.D.W.-only commenters.
- Recommendation simulations show Alt-lite content is easily reachable from I.D.W. channels, while Alt-right videos are reachable only through channel recommendations: channel reachability@5 for Alt-right was about 4% from Alt-lite and 1.5% from I.D.W.; starting from I.D.W., there was about a 10% chance of being in an Alt-lite channel at the next step and 25% chance of having found at least one Alt-lite channel in 5 steps. Video recommendations made Alt-right much less reachable, with reachability@5 around 0.05% from Alt-lite.
Methods
- Large-scale audit of YouTube channels classified into Media, Alt-lite, I.D.W., and Alt-right. Data collection used seed channels, keywords, and iterative featured/related-channel expansion; two authors manually annotated channels over 3 weeks with interannotator agreement of 75.57% (95% CI [67.5, 82.5]), yielding 85 I.D.W., 112 Alt-lite, 84 Alt-right, and 68 media channels.
- Dataset: 349 channels, 330,925 videos, 72,069,878 comments, 5,980,709 commenting users, 2,474,044 video recommendations, and 14,283 channel recommendations. Data collection occurred 19–30 May 2019, with recommendations collected May–July 2019 using crawlers, 22 channel-recommendation rounds, 19 video-recommendation rounds, VPNs from 7 locations, random channel visit order, and no logged-in personalization.
- Analyses included Jaccard and Overlap Coefficient set similarities across commenting-user sets, cohort tracking of users who commented only on milder communities and later commented on Alt-right videos, and random-walk simulations on weighted recommendation graphs (10K runs, 5 steps) to measure probability and reachability.
Limitations
- The paper states that it cannot obtain the entire history of YouTube recommendations, so algorithmic analyses are limited to a time slice of a constantly changing black box; it also cannot fully tease apart the algorithm’s influence in earlier periods, and the recommendation analysis does not account for personalization.
- The paper states that it provides little insight into why radicalization pipelines exist, and that its focus is on basic statistics and user trajectories rather than tracing the evolution of speech or narratives over time.
- The paper states that its radicalization findings rely on the assumptions that commenting users are a good enough proxy for radicalization and that comments are supportive of the videos they accompany; it supports these assumptions with a manual check of 900 randomly selected comments, finding only 5 that could be interpreted as criticisms, and with higher like proportions for the communities of interest (>91% mean, >96% median) than for media channels (85% mean, 93% median). Evident from the text: channel labeling is fuzzy and conservative, and the video-recommendation simulation is less realistic because it ignores untracked videos.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
75 Own result · 18 Own method · 12 Own interpretation · 10 Not checked yet · 7 No basis given
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This effect is much stronger than for the large traditional and alternative media channels we collected (Sec. 6).”
boostedcorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“and that Alt-right channels may be reached from the two other communities (Sec. 7).”
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“However, in 2018, there is a sharp growth in the similarity with media channels.”
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“as we can see that the communities of interest have many more highly active commenters.”
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Moreover, comments per view seem to be particularly high in more extreme content, reaching near to 1 comment for every 5 views in Alt-right channels in 2018 (Sec. 4).”
- Also stated in: § 4 THE RISE OF CONTRARIANS · p. 5 · plain
“Notably, Alt-right channels have, since 2017, become the ones with the highest number of comments per view, with nearly 1 comment per 5 views by 2018.”
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 4 THE RISE OF CONTRARIANS · p. 5 · plain
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We find that Alt-lite content is easily reachable from I.D.W. channels”
- Also stated in: § 1 INTRODUCTION · p. 2 · plain
“Our analyses show that, particularly through the channel recommender system, Alt-lite channels are easily discovered from I.D.W. channels”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 1 INTRODUCTION · p. 2 · plain
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Processing 72M+ comments, we show that the three channel types indeed increasingly share the same user base”
- Also stated in: § 1 INTRODUCTION · p. 2 · plain
“We inspect the intersection of commenting users across the communities, finding they increasingly share the same user base.”
- Also stated in: § 5 USER INTERSECTION · p. 6 · plain
“We again see that their users are more engaged, and, notably, find that the I.D.W, the Alt-lite, and the Alt-right increasingly share the same commenting user base.”
boostedcorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 1 INTRODUCTION · p. 2 · plain
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“that users consistently migrate from milder to more extreme content”
- Also stated in: § 1 INTRODUCTION · p. 2 · plain
“We also find that the intersection is not only growing due to new users but that there is significant user migration among the communities being studied.”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 1 INTRODUCTION · p. 2 · plain
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“These users are a significant fraction of the Alt-right commenting user base.”
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“while Alt-right videos are reachable only through channel recommendations.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- It may state two separate claims as one. 27
- The quote only partly states this claim. 2
- Probably not a substantial claim. 15
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Most of the initial videos in the channel are about working out or video-games, with politics related videos becoming increasingly occurring.”
plaincorrelationalSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“However, for media channels, we find that the Jaccard similarity is plateauing since 2014 and that the overlap coefficient only recently started to grow”
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“while media and Alt-lite channels had published nearly 40% of their content, the Alt-right had published a bit more than 20%.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Notably, in 2018, the Jaccard Similarity between the Alt-lite and the I.D.W. reached almost 30%, which is more than the self-similarity between the two communities.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Alt-lite and I.D.W. channels recommend channels from the same community around 50% of the time, and recommend each other around 14% (Alt-L to I.D.W.) and 23% (I.D.W. to Alt-L) of the time.”
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We show that this migration phenomenon is not only consistent throughout the years, but also that it is significant in its absolute quantity.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- The authors state this as an assumption: Commenting users are a good enough proxy for radicalization.
- The authors state this as an assumption: Comments in YouTube channels are supportive of the videos they are associated with.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Less than 1% of users in media channels were mildly or severely exposed, against 3% to 4% for Alt-lite or I.D.W. users, and roughly 4% were lightly exposed versus approximately 8% for Alt-lite or I.D.W users.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“So, while in 2018, 23 . 3% of users who were lightly exposed can be traced back to users who commented on Alt-lite channels in previous years, only 7 . 6% can be traced back to I.D.W. channels.”
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“More relevant, though, starting from media channels, the reachability@5 of I.D.W. and Alt-lite channels is of around 4 . 5% and 1 . 5% respectively.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This number is much more expressive than the number of users which came from media channels - in the last row - which never surpasses 6% for any level of exposure.”
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“here we see that users who initially commented only on I.D.W. channels constitute a much less significant percentage of the Alt-right consumer base in upcoming years.”
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The quote only partly states this claim.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Less than 1% of users in media channels were mildly or severely exposed, against 3% to 4% for Alt-lite or I.D.W. users, and roughly 4% were lightly exposed versus approximately 8% for Alt-lite or I.D.W users.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For channel recommendations, we have that the reachability@5 of Alt-right channels is of approximately 4% for the simulations starting from Alt-lite 1 . 5% for I.D.W. channels.”
hedgedcorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For users who were tracked in 2006 -2012, around 15% were exposed in both scenarios, while for those tracked in 2017, this difference grew farther apart ( ∼ 12% Alt-lite vs. ∼ 6% I.D.W.).”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For all levels of exposure, at all times, the number of exposed users that can be traced back to commenting exclusively on I.D.W. channels is around 3 times lower.”
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The quote only partly states this claim.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“When teasing apart users that commented only on Alt-lite or only on I.D.W. content, we find that, not only users who commented only on I.D.W. get less exposed, but increasingly less so.”
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Starting from the media channels, reachability@5 of I.D.W. channels is of 2 . 5%, and of slightly less than 1% for Alt-lite channels.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This phenomenon is significant both in terms of the percentage of the users tracked - as in Fig. 3 - and in terms of the total Alt-right commenting user base - as in Fig. 4.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Moreover, we note that the proportion of likes for the communities of interest is higher for the communities of interest ( > 91% mean, > 96% median) than for the media channels (85% mean, 93% median), which suggests the people interacting with the three communities agree with their videos.”
hedgedcorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- The authors state this as an assumption: Commenting users are a good enough proxy for radicalization.
- The authors state this as an assumption: Comments in YouTube channels are supportive of the videos they are associated with.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“From these two communities, it is possible to find Alt-right content from recommended channels, but not from recommended videos.”
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“To address this problem, in this paper we take a conservative approach to our labeling, naming only the most extreme content creators as Alt-right.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Each seed was independently annotated two times and discarded in case there was any disagreement.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“The annotation process lasted for 3 weeks. In case they disagreed, they had to discuss the cases individually until a conclusion was reached.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Although comments may give us insight into the past, it is challenging to tease apart the influence of the algorithm in previous times.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Notice that we do not have the user's entire activity history, and thus, we track their activity only in the channels whose videos we collected.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“and thus it is hard to tell whether their users are less engaged.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Moreover, channels were always visited in random order, to prevent any biases from arising from session-based recommendations.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“To circumvent possible location bias in the data we collected we used VPNs from 7 different locations: 3 in the USA, 2 in Canada, 1 in Switzerland and 1 in Brazil.”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In the video recommender system, Alt-right channels are less recommended, but finding Alt-lite channels from the I.D.W. and I.D.W. channels from the large media channels in the media group is also feasible.”
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This process may also be amplified when taking personalization into account.”
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Overall, we find that, in the channel recommender system, it is easy to navigate from the I.D.W. to the Alt-lite (and vice-versa), and it is possible to find Alt-right channels.”
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
In the paper
“We argue that this finding provides significant evidence that there has been, and there continues to be, user radicalization on YouTube, and our analyses of the activity of these communities (Sec. 4) is consistent with the theory that more extreme content 'piggybacked' on the surge in popularity of I.D.W. and Alt-lite content [30].”
What the sources say
- Not checked yetcites [30] [30] Angela Nagle. 2017. Kill All Normies: Online Culture Wars From 4Chan And Tumblr To Trump And The Alt-Right . -
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- The authors state this as an assumption: Commenting users are a good enough proxy for radicalization.
- The authors state this as an assumption: Comments in YouTube channels are supportive of the videos they are associated with.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“However, it does not give insight into how expressive the effect is in terms of what part of the Alt-right user base has gone through it.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Unfortunately, we have only a snapshot of the recommender system which does not take into account personalization. Thus, it is hard to reach significant conclusions on what was the role of the recommender system in the radicalization process we depicted in Sec. 6.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
None of the sources cited for this claim have been checked yet.
In the paper
“The previous study suggests that the pipeline effect does exist, and that indeed, users systematically go from milder communities to the Alt-right.”
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“This is, to our best knowledge, the first large scale quantitative audit of user radicalization on YouTube.”
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
- Rests on an assumption: There is to date no substantial quantitative evidence of the alleged radicalization pipeline on YouTube.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Since the channel is small, removing it has no noticeable impact on the results presented in this paper.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The removed channel 'theglassblindspot' is small.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Although we argue this is strong evidence for the existence of radicalization pathways on YouTube, our work provides little insight on why these radicalization pipelines exist.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- The authors state this as an assumption: Commenting users are a good enough proxy for radicalization.
- The authors state this as an assumption: Comments in YouTube channels are supportive of the videos they are associated with.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Notice that the scenario for the channels is more realistic, and we give more weight to the conclusions drawn there.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Noticeably, our analysis has several shortcomings which do not allow us to make bold claims about this research question. Firstly, we are able to look only at a tiny fraction of actual recommendations - it could very well be that Alt-right content was being more widely promoted in the past. Secondly, our analysis does not take into account personalization, which could reveal a completely different”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper relies on this without support.
There is to date no substantial quantitative evidence of the alleged radicalization pipeline on YouTube.
“Yet, there is to date no substantial quantitative evidence of this alleged pipeline.”
The authors state this as an assumption.
Commenting users are a good enough proxy for radicalization.
“Noticeably, the findings related to this research question make the implicit assumption that commenting users are a good enough proxy for radicalization, and that comments in YouTube channels are supportive of the videos they are associated with.”
The authors state this as an assumption.
Comments in YouTube channels are supportive of the videos they are associated with.
“Noticeably, the findings related to this research question make the implicit assumption that commenting users are a good enough proxy for radicalization, and that comments in YouTube channels are supportive of the videos they are associated with.”
The paper relies on this without support.
The removed channel 'theglassblindspot' is small.
“Since the channel is small”
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“Such path dependencies may have transformative effects, changing the preferences and values of the users themselves and leading to radicalization”
From Stewardship of global collective behavior (2021) · cites this paper as [128]
In the citing paper
“Despite the opacity, research has revealed how algorithms lead users to radical or age-inappropriate content (128, 164), exacerbate disparities in health (123), and increase bias in policing (124).”
From Stewardship of global collective behavior (2021) · cites this paper as [128]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
“| Alt-R | 25.73/ 1.89 | 42.94/ 1.15 | 25.73/ 8.55 | 1.35/ 3.38 | 21.08/ 85.03 |”
Reported 100 · recomputed 116.83
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 7 THE RECOMMENDATION ALGORITHM · p. 9
“Alt-right channels are only recommended by Alt-lite channels (3 . 08%).”
§ 7 THE RECOMMENDATION ALGORITHM · p. 8
“| Alt-L | 13.69/ 2.46 | 55.15/ 12.70 | 3.38/ 0.13 | 2.82/ 3.24 | 24.96/ 81.47 |”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ ABSTRACT · p. 1
“Alt-right videos are reachable only through channel recommendations.”
§ 7 THE RECOMMENDATION ALGORITHM · p. 10
“For video recommendations, reaching Alt-right channels from other communities is less likely. From the Alt-lite, reachability@5 is of around 0 . 05%.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
“| | | 2017 | 2962 | 5691 | 0.64% | 525% |”
Reported 525 · recomputed 52.047092
Citations
Cites
Deep Neural Networks for YouTube Recommendations
Covington, P., Adams, J., & Sargin, E. (2016) · Proceedings of the 10th ACM Conference on Recommender Systems · doi:10.1145/2959100.2959190 · Cited by 3,539
The YouTube video recommendation system
Davidson, J., Liebald, B., Liu, J., Nandy, P., Vleet, T. V., Gargi, U., Gupta, S. K., He, Y. H., Lambert, M., Livingston, B., et al. (2010) · ACM Conference on Recommender Systems (RecSys) · doi:10.1145/1864708.1864770 · Cited by 1,282
Mechanisms of Political Radicalization: Pathways Toward Terrorism
McCauley, C. R., & Moskalenko, S. (2008) · Terrorism and Political Violence · doi:10.1080/09546550802073367 · Cited by 1,222
(2018) · Cited by 900
Digital Discrimination: The Case of Airbnb.com
Edelman, B. G., & Luca, M. (2014) · doi:10.2139/ssrn.2377353 · Cited by 471
Measuring price discrimination and steering on e-commerce web sites
Hannak, A., Soeller, G., Lazer, D., Mislove, A., & Wilson, C. (2014) · Proceedings of the 2014 Conference on Internet Measurement Conference · doi:10.1145/2663716.2663744 · Cited by 405 · cited 1×
“Previous large scale algorithmic auditing include measuring discrimination on AirBnB [14], personalization onweb search [19] and price discrimination on e-commerce web sites [20].”
BackgroundMeasuring personalization of web search
Hannak, A., Sapiezynski, P., Molavi Kakhki, A., Krishnamurthy, B., Lazer, D., Mislove, A., & Wilson, C. (2013) · Proceedings of the 22nd international conference on World Wide Web · doi:10.1145/2488388.2488435 · Cited by 396
Kill All Normies: Online Culture Wars from 4chan and Tumblr to Trump and the Alt-Right
Nagle, A. (2017) · Cited by 306
Alternative Influence: Broadcasting the Reactionary Right on YouTube
Lewis, R. A. (2018) · doi:10.69985/snfr4363 · Cited by 303
On the Origins of Memes by Means of Fringe Web Communities
Zannettou, S., Caulfield, T., Blackburn, J., Cristofaro, E. D., Sirivianos, M., Stringhini, G., & Suárez‐Tangil, G. (2018) · ACM Internet Measurement Conference (IMC) · doi:10.1145/3278532.3278550 · Cited by 296
Down the (White) Rabbit Hole: The Extreme Right and Online Recommender Systems
O’Callaghan, D., Greene, D., Conway, M., Carthy, J., & Cunningham, P. (2014) · Social Science Computer Review · doi:10.1177/0894439314555329 · Cited by 245
Algorithmic Accountability Reporting: On the Investigation of Black Boxes
Diakopoulos, N. A. (2014) · Columbia Academic Commons (Columbia University) · doi:10.7916/d8zk5tw2 · Cited by 222
Zannettou, S., Caulfield, T., Cristofaro, E. D., Kourtelris, N., Leontiadis, I., Sirivianos, M., Stringhini, G., & Blackburn, J. (2017) · ACM Internet Measurement Conference (IMC) · doi:10.1145/3131365.3131390 · Cited by 209
Sellars, A. (2016) · SSRN Electronic Journal · doi:10.2139/ssrn.2882244 · Cited by 130
Analyzing Right-wing YouTube Channels
Ottoni, R., Cunha, E. L. T. P., Magno, G., Bernardina, P., Meira, W., & Almeida, V. A. F. (2018) · ACM Web Science Conference (WebSci) · doi:10.1145/3201064.3201081 · Cited by 70
Self-citationMining YouTube to Discover Extremist Videos, Users and Hidden Communities
Sureka, A., Kumaraguru, P., Goyal, A., & Chhabra, S. (2010) · Lecture notes in computer science · doi:10.1007/978-3-642-17187-1_2 · Cited by 64
A focused crawler for mining hate and extremism promoting videos on YouTube.
Agarwal, S., & Sureka, A. (2014) · ACM Conference on Hypertext and Social Media (HT) · doi:10.1145/2631775.2631776 · Cited by 61
Charlottesville and the alt-right: a turning point?
Atkinson, D. C. (2018) · Politics Groups and Identities · doi:10.1080/21565503.2018.1454330 · Cited by 50
Analyzing Right-wing YouTube Channels: Hate, Violence and DiscriminationPreprint
Ottoni, R., Cunha, E. L. T. P., Magno, G., Bernadina, P., Meira, W., & Almeida, V. A. F. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1804.04096 · Cited by 42
Self-citationA Multimodal Approach to Violence Detection in Video Sharing Sites
Γιαννακόπουλος, Θ., Pikrakis, A., & Theodoridis, S. (2010) · Proceedings - International Conference on Pattern Recognition/Proceedings/International Conference on Pattern Recognition · doi:10.1109/icpr.2010.793 · Cited by 38
Papadamou, K., Papasavva, A., Zannettou, S., Blackburn, J., Kourtellis, N., Leontiadis, I., Stringhini, G., & Sirivianos, M. (2019) · OpenBU (Boston University) · doi:10.5281/zenodo.3739061 · Cited by 20
Papadamou, K., Papasavva, A., Zannettou, S., et al. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1901.07046 · Cited by 2
Research works printed but not linked: 4
- Mathew Ingram . 2018. Most Americans say they have lost trust in the media. Columbia Journalism Review ( 2018 ). https://www.cjr.org/the_media_today/trust-in-media-down.php Mathew Ingram. 2018. Most Americans say they have lost trust in the media. Columbia Journalism Review (2018). https://www.cjr.org/the_media_today/trust-in-media-down.php
- Brendan Joel Kelly . 2017 . Lauren Southern: The alt-right's Canadian dog whistler. https://www.splcenter.org/hatewatch/2017/11/07/lauren-southern-alt-right%E2%80%99s-canadian-dog-whistler Brendan Joel Kelly. 2017. Lauren Southern: The alt-right's Canadian dog whistler. https://www.splcenter.org/hatewatch/2017/11/07/lauren-southern-alt-right%E2%80%99s-canadian-dog-whistler
- Max Kutner . 2016. Roosh V's journey from pickup artist to right-wing provocateur. Newsweek ( 2016 ). https://www.newsweek.com/2016/10/21/roosh-v-pickup-artist-right-wing-provocateur-509319.html Max Kutner. 2016. Roosh V's journey from pickup artist to right-wing provocateur. Newsweek (2016). https://www.newsweek.com/2016/10/21/roosh-v-pickup-artist-right-wing-provocateur-509319.html
- Newman Nic , Richard Fletcher , Antonis Kalogeropoulos , David AL Levy, and Rasmus Kleis Nielsen . 2018 . Reuters Institute Digital News Report 2018. Technical Report. Reuters Institute for the Study of Journalism. Newman Nic, Richard Fletcher, Antonis Kalogeropoulos, David AL Levy, and Rasmus Kleis Nielsen. 2018. Reuters Institute Digital News Report 2018. Technical Report. Reuters Institute for the Study of Journalism.
Cited by
Sorted by citations
A systematic review of worldwide causal and correlational evidence on digital media and democracy
Lorenz-Spreen, P., Oswald, L., Lewandowsky, S., & Hertwig, R. (2022) · Nature Human Behaviour · doi:10.1038/s41562-022-01460-1 · Cited by 348
Stewardship of global collective behavior
Bak-Coleman, J. B., Alfano, M., Barfuss, W., Bergstrom, C. T., Centeno, M. A., Couzin, I. D., Donges, J. F., Galesic, M., Gersick, A. S., Jacquet, J., et al. (2021) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2025764118 · Cited by 289
False equivalencies: Online activism from left to right
Freelon, D., Marwick, A., & Kreiss, D. (2020) · Science · doi:10.1126/science.abb2428 · Cited by 281
Right-Wing YouTube: A Supply and Demand Perspective
Munger, K., & Phillips, J. (2022) · The International Journal of Press/Politics · doi:10.1177/1940161220964767 · Cited by 265
Dancing to the Partisan Beat: A First Analysis of Political Communication on TikTok
Serrano, J. C. M., Papakyriakopoulos, O., & Hegelich, S. (2020) · ACM Web Science Conference (WebSci) · doi:10.1145/3394231.3397916 · Cited by 264
Metadata
- Authors
- Ribeiro, M. H., Ottoni, R., West, R., Almeida, V. A. F., & Meira, W.
- Affiliations
- École Polytechnique Fédérale de Lausanne (CH)
- Universidade Federal de Minas Gerais (BR)
- UFMG
- Venue
- Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency
- Publisher
- ACM
- Published
- 2020-01-27
- Pages
- 131-141
- OpenAlex
- W2969399596
- Semantic Scholar
- 57911e835999c5b0fc8aea382bd7569a180503ae
- Cited by
- 580
- References
- 23
- Citations by year
- 2026: 110 · 2025: 70 · 2024: 89 · 2023: 105 · 2022: 78 · 2021: 69 · 2020: 52
- FWCI
- 161.71
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- ACM Conference on Fairness, Accountability, and Transparency (FAccT) (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- Infoscience (Ecole Polytechnique Fédérale de Lausanne) (submittedVersion, cc-by-nc-nd)
- arXiv (Cornell University)
- Funders
- Atmosphere
- Capes
- INCT-Cyber
- Google Research Award
- Fapemig
- CNPq
- MASWEB
- Topics
- Social Sciences › Sociology and Political Science › Misinformation and Its Impacts
- Social Sciences › Political Science and International Relations › Populism, Right-Wing Movements
- Social Sciences › Communication › Social Media and Politics
- Keywords
- radicalization, YouTube, far-right ideology, alt-right, user engagement
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
A language model drew the claims from the paper on 2026-10-09, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-09 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
