standOff. When we have changed our mind. Some entries in the table begin with a word in capitals and the reason for it. EVIDENCE CORRECTED — the finding still stands, but we had quoted the wrong sentence from the paper, and the quotation has been replaced with the right one. RETRACTED — the entry itself was wrong; it stays on the page, marked, with the reason, rather than being deleted. WITHDRAWN AND REPLACED — the entry was wrong, and a corrected version of it appears in the entries below. Each mark carries its own reason, in brackets, because the reasons differ from one entry to the next. An entry we have WITHDRAWN is removed from this page and kept in the record, with its reason. A line reading Audit and a date is a later re-reading of the entry against its whole sentence: DOUBTFUL means it was kept, and the reason says what is loose about it. Quotation marks inside the tagger's sentence. Where a tag is the tagger's own sentence, a phrase in quotation marks inside it is borrowed from the paper — the tagger's wording around it is not. The paper's own sentence is the evidence quote shown on the same row, so the borrowing can be checked against the thing borrowed from without leaving this page. Nothing inside quotation marks anywhere on this page is ours: if it is quoted, it is the paper's. Whose words a concept carries. A concept entry records a term the paper itself uses, so the term, the sense given for it, and its evidence are the paper's own words, quoted where the paper fixes them. This is the opposite of a sense entry in the tag table, which is the tagger's sentence about how the paper uses a term. Where a concept entry could not be grounded in the paper's words, the entry says so on its own row.| claim type | claim_type | the shape of the claim: existence, mechanism, method, critique, prediction, null |
| concept edge | concept_edges | an asserted relation between two concepts, with the sentence asserting it |
| concept | concepts | an idea the paper works with, quoted where the paper uses it |
| content tag | content_tags | a finding, definition, quantity, statistic or variable named in the paper |
| contribution | contribution | what the paper claims to add |
| control | controls | an experiment run to eliminate a rival explanation |
| limitation | limitations | what the paper says its evidence cannot support |
| mechanism | mechanisms | how the reported effect is said to come about |
| method | method | how the paper produced its evidence |
| object of study | object_of_study | what the paper is about — the thing it investigates |
| phenomenon | phenomenon | the effect or regularity under study |
| research question | research_question | the question the paper sets itself |
| sense | senses | a distinct use of a term, quoted where the paper fixes it |
Users Polarization on Facebook and Youtube
Abstract
concepts↗content_tagsUsers online tend to select information that support and adhere their beliefs, and to form polarized groups sharing the same view—e.g. echo chambers. Algorithms for content promotion may favour this phenomenon, by accounting for users preferences and thus limiting the exposure to unsolicited contents. To shade light on this question, we perform a comparative study on how same contents (videos) are consumed on different online social media—i.e. Facebook and YouTube—over a sample of 12 M of users. Our findings show that content drives the emergence of echo chambers on both platforms. Moreover, we show that the users’ commenting patterns are accurate predictors for the formation of echo-chambers.
Introduction
mechanismsconcepts↗sensescontent_tags↗content_tags↗The diffusion of social media caused a shift of paradigm in the creation and consumption of information. We passed from a mediated (e.g., by journalists) to a more disintermediated selection process. Such a disintermediation elicits the tendencies of the users to a) select information adhering to their system of beliefs—i.e., confirmation bias—and b) to form groups of like minded people where they polarize their view—i.e. echo chambers [ 1 – 6 ]. Polarized communities emerge around diverse and heteorgeneous narratives often reflecting extreme disagreement with respect to the main stream news and recommended practices. The emergence of polarization in online environments might reduce viewpoint heterogeneity, which has long been viewed as an important component of democratic societies [ 7 , 8 ].
mechanismsmechanismslimitationsconcepts↗concept_edgesConfirmation bias has been shown to play a pivotal role in the diffusion of rumors online [ 9 ]. However, on online social media, different algorithms foster personalized contents according to user tastes—i.e. they show users viewpoints that they already agree with. The role of these algorithms in influencing the emergence of echo chambers is still a matter of debate. Indeed, little is known about the factors affecting the algorithms’ outcomes. Facebook promotes posts according to the News Feed algorithm, that helps users to see more stories from friends they interact with the most, and the number of comments and likes a post receives and what kind of story it is—e.g. photo, video, status update—can also make a post more likely to appear [ 10 ]. Conversely, YouTube promotes videos through Watch Time , which prioritizes videos that lead to a longer overall viewing session over those that receive more clicks [ 11 ]. Not much is known about the role of cognitive factors in driving users to aggregate in echo chambers supporting their preferred narrative. Recent studies suggest confirmation bias as one of the driving forces of content selection, which eventually leads to the emergence of polarized communities where users acquire confirmatory information and ignore dissenting content [ 12 – 17 ].
research_questioncontrolsTo shade light on the role of algorithms for content promotion in the emergence of echo chambers, we analyze the users behavior exposed to the same contents on different platforms—i.e. Youtube and Facebook. We focus on Facebook posts linking Youtube videos reported on Science and Conspiracy pages. We then compare the users interaction with these videos on both platforms.
limitationscontrolsconcept_edgesWe limit our analysis to Science and Conspiracy for two main reasons: a) scientific news and conspiracy-like news are two very distinct and conflicting narratives; b) scientific pages share the main mission to diffuse scientific knowledge and rational thinking, while the alternative ones resort to unsubstantiated rumors.
Indeed, conspiracy-like pages disseminate myth narratives and controversial information, usually lacking supporting evidence and most often contradictory of the official news. Moreover, the spreading of misinformation on online social media has become a widespread phenomenon to an extent that the World Economic Forum listed massive digital misinformation as one of the main threats for the modern society [ 16 , 18 ].
In spite of different debunking strategies, unsubstantiated rumors—e.g. those supporting anti-vaccines claims, climate change denials, and alternative medicine myths—keep proliferating in polarized communities emerging on online environments [ 9 , 14 ], leading to a climate of disengagement from mainstream society and recommended practices. A recent study [ 19 ] pointed out the inefficacy of debunking and the concrete risk of a backfire effect [ 20 , 21 ] from the usual and most committed consumers of conspiracy-like narratives.
We believe that additional insights about cognitive factors and behavioral patterns driving the emergence of polarized environments are crucial to understand and develop strategies to mitigate the spreading of online misinformation.
object_of_studymethod↗In this paper, using a quantitative analysis on a massive dataset (12 M of users), we compare consumption patterns of videos supporting scientific and conspiracy-like news on Facebook and Youtube. We extend our analysis by investigating the polarization dynamics—i.e. how users become polarized comment after comment. On both platforms, we observe that some users interact only with a specific kind of content since the beginning, whereas others start their commenting activity by switching between contents supporting different narratives. The vast majority of the latter—after the initial switching phase—starts consuming mainly one type of information, becoming polarized towards one of the two conflicting narratives. Finally, by means of a multinomial logistic model, we are able to predict with a good precision the probability of whether a user will become polarized towards a given narrative or she will continue to switch between information supporting competing narratives. The observed evolution of polarization is similar between Facebook and YouTube to an extent that the statistical learning model trained on Facebook is able to predict with a good precision the polarization of YouTube users, and vice versa. Our findings show that contents more than the algorithms lead to the aggregation of users in different echo chambers.
Results and Discussion
concepts↗concepts↗sensescontent_tags↗content_tags↗content_tags↗content_tags↗content_tags↗We start our analysis by focusing on the statistical signatures of content consumption on Facebook and Youtube videos. The focus is on all videos posted by conspiracy-like and scientific pages on Facebook. We compare the consumption patterns of the same video on both Facebook and Youtube. On Facebook a like stands for a positive feedback to the post; a share expresses the will to increase the visibility of a given information; and a comment is the way in which online collective debates take form around the topic promoted by posts. Similarly, on YouTube a like stands for a positive feedback to the video; and a comment is the way in which online collective debates grow around the topic promoted by videos.
Contents Consumption across Facebook and YouTube
content_tagsAs a preliminary analysis we measure the similarity of the users reaction to the same videos on both platforms. Focusing on the consumptions patterns of YouTube videos posted on Facebook pages, we compute the Spearman’s rank correlation coefficients between users’ actions on Facebook posts and the related YouTube videos (see Fig 1 ). We find strong correlations on how users like, comment and share videos on Facebook and Youtube. Despite the different algorithm for content promotion, information reverberate in a similar way.
Correlation Matrix. Spearman’s rank correlation coefficients between users’ actions on Facebook posts and the related YouTube videos.
sensescontent_tagscontent_tagsBy means of the Mantel test [ 22 ] we find a statistically significant (simulated p-value <0.01, based on 10 4 Monte Carlo replicates), high, and positive ( r = 0.987) correlation between the correlation matrices of Science and Conspiracy. In particular, we find positive and high correlations between users’ actions on YouTube videos for both Science and Conspiracy, indicating a similar strong monotone increasing relationship between views, likes, and comments. Furthermore, we observe positive and mild correlations between users’ actions on Facebook posts linking YouTube videos for both Science and Conspiracy, suggesting a monotone increasing relationship between likes, comments, and shares. Conversely, we find positive yet low correlations between users’ actions across YouTube videos and the Facebook posts linking the videos for both Science and Conspiracy, implying that the success—in terms of received attention—of videos posted on YouTube does not ensure a comparable success on Facebook, and vice versa. This evidence suggests that the social response to information is similar on different contents and platforms.
As a further analysis we focus on the volume of actions to each post. In Fig 2 we show the empirical Cumulative Complementary Distribution Functions (CCDFs) of the consumption patterns of videos supporting conflicting narratives—i.e. Science and Conspiracy—in terms of comments and likes on Facebook and YouTube. The double-log scale plots highlight the power law behavior of each distribution. Top right panel shows the CCDFs of the number of likes received by Science ( x min = 197 and θ = 1.96) and Conspiracy ( x min = 81 and θ = 1.91) on Facebook. Top left panel shows the CCDFs of the number of comments received by Science ( x min = 35 and θ = 2.37) and Conspiracy ( x min = 22 and θ = 2.23) on Facebook. Bottom right panel shows the CCDFs of the number of likes received by Science ( x min = 1,609 and θ = 1.65) and Conspiracy ( x min = 1,175 and θ = 1.75) on YouTube. Bottom left panel shows the CCDFs of the number of comments received by Science ( x min = 666 and θ = 1.70) and Conspiracy ( x min = 629 and θ = 1.77) on YouTube.
Consumption Patterns of Videos on Facebook and YouTube. The empirical CCDFs, 1 − F ( x ), show the consumption patterns of videos supporting conflicting narratives—i.e. Science and Conspiracy—in terms of comments (A and C) and likes (B and D) on Facebook and YouTube.
content_tagsSocial response on different contents do not present a significant difference on Facebook and Youtube. Users’ response to content is similar on both platform and on both types of content. Science and Conspiracy videos receive the same amount of attention and reverberate in a similar way.
Polarized and Homogeneous Communities
content_tagscontent_tagscontent_tagsAs a secondary analysis we want to check whether the content has a polarizing effect on user. Hence, we focus on the users’ activity across the different type of contents. Fig 3 shows the Probability Density Functions (PDFs) of about 12 M users’ and on how they distribute their comments on Science and Conspiracy posts (polarization) on both Facebook and YouTube. We observe sharply peaked bimodal distributions. Users concentrate their activity on one of the two narratives. To quantify the degree of polarization we use the Bimodality Coefficient (BC), and we find that the BC is very high for both Facebook and YouTube. In particular, BC FB = 0.964 and BC YT = 0.928. Moreover, we observe that the percentage of polarized users (users with ρ < 0.05 and ρ > 0.95) is 93.6% on Facebook and 87.8% on YouTube; therefore, two well separated communities support competing narratives in both online social networks.
Polarization on Facebook and YouTube. The PDFs of the polarization ρ show that the vast majority of users is polarized towards one of the two conflicting narratives—i.e. Science and Conspiracy—on both Facebook and YouTube.
concept_edgescontent_tagsContent has a polarizing effect, indeed, users focus on specific types of content and aggregate in separated groups—echo chambers—independently of the platform and content promotion algorithm.
content_tags↗To further detail such a segregation, we analyze how polarized users—i.e., users having more than the 95% of their interactions with one narrative—behave with respect to their preferred content. Fig 4 shows the empirical CCDFs of the number of comments left by all polarized users on Facebook and YouTube ( x m i n F B = 8 , θ FB = 2.13 and x m i n Y T = 17 , θ YT = 2.29). We observe a very narrow difference ( HDI 90 = [−0.18,−0.13]) between the tail behavior of the two distributions. Moreover, Fig 5 shows the empirical CCDFs of the number of comments left by users polarized on either Science or Conspiracy on both Facebook ( x m i n S c i = 5 , θ Sci = 2.29 and x m i n C o n = 4 , θ Con = 2.31, with HDI 90 = [−0.018,−0.009]) and YouTube ( x m i n S c i = 2 , θ Sci = 2.86 and x m i n C o n = 3 , θ Con = 2.41, with HDI 90 = [0.44, 0.46]). Users supporting conflicting narratives behave similarly on Facebook, whereas on YouTube the power law distributions slightly differ in the scaling parameters.
Commenting Activity of Polarized Users. The empirical CCDFs, 1 − F ( x ), of the number of comments left by polarized users on Facebook and YouTube.
Commenting Activity of Users Polarized towards Conflicting Narratives. The empirical CCDFs, 1 − F ( x ), of the number of comments left by users polarized on scientific narratives and conspiracy theories on Facebook (A) and YouTube (B).
phenomenonconcepts↗content_tagsThe aggregation of users around conflicting narratives lead to the emergence of echo chambers. Once inside such homogeneous and polarized communities, users supporting both narratives behave in a similar way, irrespective of the platform and content promotion algorithm.
Prediction of Users Polarization
sensescontent_tagscontent_tags↗content_tags↗Now we want to characterize how the content attract users,—i.e. how users’ polarization evolves comment after comment. We consider random samples of 400 users who left at least 100 comments, and we compute the mobility of a user across different contents along time. On both Facebook and YouTube, we observe that some users interact with a specific kind of content, whereas others start their commenting activity by switching between contents supporting different narratives. The latter—after an initial switching phase—starts focusing only on one type of information, becoming polarized towards one of the two conflicting narratives. We exploit such a regularity to derive a data-driven model to forecast users’ polarizations. Indeed, by means of a multinomial logistic model, we are able to predict the probability of whether a user will become polarized towards a given narrative or she will continue to switch between information supporting competing narratives. In particular, we consider the users’ polarization after n comments, ρ n with n = 1, …, 100, as a predictor to classify users in three different classes: Polarized in Science ( N = 400), Not Polarized ( N = 400), Polarized in Conspiracy ( N = 400).
controlscontrolscontent_tagscontent_tagsFig 6 shows precision, recall, and accuracy of the classification tasks on Facebook and YouTube as a function of n . On both online social networks, we find that the model’s performances monotonically increase as a function of n for each class. Focusing on accuracy, significant results (greater than 0.70) are obtained for low values of n . A suitable compromise between classification performances and required number of comments seems to be n = 50, which provides an accuracy greater than 0.80 for each class on both YouTube and Facebook. To assess how the results generalize to independent datasets and to limit problems like overfitting, we split YouTube and Facebook users datasets in training sets ( N = 1000) and test sets ( N = 200), and we perform Monte Carlo cross validations with 10 3 iterations. Results of Monte Carlo validations are shown in Table 1 and confirm the goodness of the model.
Performance measures the classification task. Precision, recall, and accuracy of the classification task for users Polarized in Conspiracy, Not Polarized, Polarized in Science on Facebook and YouTube as a function of n . On both online social networks, we find that the model’s performance measures monotonically increase as a function of n . Focusing on the accuracy, significant results (greater than 0.70) are obtained for low values of n .
Monte Carlo Cross Validation. Mean and standard deviation (obtained averaging results of 10 3 iterations) of precision, recall, and accuracy of the classification task for users Polarized in Conspiracy, Not Polarized, Polarized in Science.
contributionmechanismscontrolsconcepts↗content_tagscontent_tagsWe conclude that the early interaction of users with contents is an accurate predictor for the preferential attachment to a community and thus for the emergence of echo chambers. Moreover, in Table 2 , we show that the evolution of the polarization on Facebook and YouTube is so alike that the same model (with n = 50), when trained with Facebook users ( N = 1200) to classify YouTube users ( N = 1200), leads to an accuracy in the classification task greater than 0.80 for each class. Similarly, using YouTube users as training set to classify Facebook users leads to similar performances.
Performance measures of classification. Precision, recall, and accuracy of the classification task for users Polarized in Conspiracy, Not Polarized, Polarized in Science when YouTube users are used as training set to classify Facebook users (top table), and when Facebook users are used as training set to classify YouTube users (bottom table).
Conclusions
Algorithms for content promotion are supposed to be the main determinants of the polarization effect arising out of online social media. Still, not much is known about the role of cognitive factors in driving users to aggregate in echo chambers supporting their favorite narrative. Recent studies suggest confirmation bias as one of the driving forces of content selection, which eventually leads to the emergence of polarized communities [ 12 – 15 ].
claim_typemechanismscontent_tagsOur findings show that conflicting narratives lead to the aggregation of users in homogeneous echo chambers, irrespective of the online social network and the algorithm of content promotion.
limitationsIndeed, in this work, we characterize the behavioral patterns of users dealing with the same contents, but different mechanisms of content promotion. In particular, we investigate whether different mechanisms regulating content promotion in Facebook and Youtube lead to the emergence of homogeneous echo chambers.
We study how users interact with two very distinct and conflicting narratives—i.e. conspiracy-like and scientific news—on Facebook and YouTube. Using extensive quantitative analysis, we find the emergence of polarized and homogeneous communities supporting competing narratives that behave similarly on both online social networks. Moreover, we analyze the evolution of polarization, i.e. how users become polarized towards a narrative. Still, we observe strong similarities between behavioral patterns of users supporting conflicting narratives on different online social networks.
Such a common behavior allows us to derive a statistical learning model to predict with a good precision whether a user will become polarized towards a certain narrative or she will continue to switch between contents supporting different narratives. Finally, we observe that the behavioral patterns are so similar in Facebook and YouTube that we are able to predict with a good precision the polarization of Facebook users by training the model with YouTube users, and vice versa.
Methods
Ethics Statement
limitationscontrolsThe entire data collection process has been carried out exclusively through the Facebook Graph API [ 23 ] and the YouTube Data API [ 24 ], which are both publicly available, and for the analysis we used only public available data (users with privacy restrictions are not included in the dataset). The pages from which we download data are public Facebook and YouTube entities. User content contributing to such entities is also public unless the user’s privacy settings specify otherwise and in that case it is not available to us. We abided by the terms, conditions, and privacy policies of the websites (Facebook/Youtube)
Data Collection
limitationsconcepts↗concepts↗sensescontent_tagscontent_tagsThe Facebook dataset is composed of 413 US public pages divided to Conspiracy and Science news. The first category (Conspiracy) includes pages diffusing alternative information sources and myth narratives—pages which disseminate controversial information, usually lacking supporting evidence and most often contradictory of the official news. The second category (Science) includes scientific institutions and scientific press having the main mission of diffusing scientific knowledge. Such a space of investigation is defined with the same approach as in [ 19 ], with the support of different Facebook groups very active in monitoring the conspiracy narratives. Pages were accurately selected and verified according to their self description. For both the categories of pages we downloaded all the posts (and their respective users interactions) in a timespan of 5 years (Jan 2010 to Dec 2014). To our knowledge, the final dataset is the complete set of all scientific and conspiracy-like information sources active in the US Facebook scenario up to date.
limitationscontent_tagsWe pick all posts on Facebook linking a video on Youtube and then through the API we downloaded the videos related metadata. To build the Youtube database of video we downloaded likes, comments and descriptions of each video cited/shared in Facebook posts using the Youtube Data API [ 25 ]. Each video link in Facebook contains an unique id that identify the resource in a unique way on both Facebook and Youtube. The comments thread in Youtube, with its time sequence, is the equivalent of the feed timeline in a Facebook page. The techniques used to analyse Facebook data can be then used in Youtube data with minimum modifications. The YouTube dataset is composed of about 17 K videos linked by Facebook posts supporting Science or Conspiracy news. Videos linked by posts in Science pages are considered as videos disseminating scientific knowledge, whereas videos linked by posts in Conspiracy pages are considered as videos diffusing controversial information and supporting myth and conspiracy-like theories. Such a categorization is validated by all the authors and Facebook groups very active in monitoring conspiracy narratives. The exact breakdown of the data is shown in Tables 3 , 4 , 5 and 6 . Summarizing, the dataset is composed of all public videos posted by the Facebook pages listed in the Page List section and their related instances on Youtube.
Breakdown of the dataset.
Conspiracy Pages.
Science Pages.
Debunking Pages.
Preliminaries and Definitions
Polarization of Users
concepts↗concept_edgescontent_tags↗content_tags↗Polarization of users, ρ u ∈ [0, 1], is defined as the fraction of comments that a user u left on posts (videos) supporting conspiracy-like narratives on Facebook (YouTube). In mathematical terms, given s u , the number of comments left on Science posts by user u , and c u , the number of comments left on Conspiracy posts by user u , the polarization of u is defined as
We then consider users with ρ u > 0.95 as users polarized towards Conspiracy, and users with ρ u < 0.05 as users polarized towards Science.
Bimodality Coefficient
The Bimodality Coefficient (BC) [ 26 ] is defined as
with μ 3 referring to the skewness of the distribution and μ 4 referring to its excess kurtosis, with both moments being corrected for sample bias using the sample size n .
content_tags↗The BC of a given empirical distribution is then compared to a benchmark value of BC crit = 5/9 ≈ 0.555 that would be expected for a uniform distribution; higher values point towards bimodality, whereas lower values point toward unimodality.
Multinomial Logistic Model
concepts↗content_tags↗Multinomial logistic regression is a classification method that generalizes logistic regression to multi-class problems, i.e. with more than two possible discrete outcomes [ 27 ]. Such a model is used to predict the probabilities of the different possible outcomes of a categorically distributed dependent variable, given a set of independent variables. In the multinomial logistic model we assume that the log-odds of each response follow a linear model
where α j is a constant and β j is a vector of regression coefficients, for j = 1, 2, …, J − 1. Such a model is analogous to a logistic regression model, except that the probability distribution of the response is multinomial instead of binomial, and we have J − 1 equations instead of one. The J − 1 multinomial logistic equations contrast each of categories j = 1, 2, …, J − 1 with the baseline category J . If J = 2 the multinomial logistic model reduces to the simple logistic regression model.
The multinomial logistic model may also be written in terms of the original probabilities π ij rather than the log-odds. Indeed, assuming that η iJ = 0, we can write
Classification Performance Measures
To assess the goodness of our model we use three different measures of classification performance: precision, recall, and accuracy. For each class i , we compute the number of true positive cases TP i , true negative cases TN i , false positive cases FP i , and false negative cases FN i . Then, for each class i the precision of the classification is defined as
the recall is defined as
and the accuracy is defined as
Power law distributions
Scaling exponents of power law distributions are estimated via maximum likelihood (ML) as shown in [ 28 ]. To provide a full probabilistic assessment about whether two distributions are similar, we estimate the posterior distribution of the difference between the scaling exponents through an Empirical Bayes method.
Suppose we have two samples of observations, A and B , following power law distributions. For the sample A , we use the ML estimate of the scaling parameter, θ ^ A M L , as location hyper-parameter of a Normal distribution with scale hyper-parameter σ ^ A M L . Such a Normal distribution represents the prior distribution, p ( θ A ) ∼ N ( θ ^ A M L , σ ^ A M L ) , of the scaling exponent θ A . Then, according to the Bayesian paradigm, the prior distribution, p ( θ A ), is updated into a posterior distribution, p ( θ A | x A ):
where p ( x A | θ A ) is the likelihood. The posterior distribution is obtained via Metropolis-Hastings algorithm, i.e. a Markov Chain Monte Carlo (MCMC) method used to obtain a sequence of random samples from a probability distribution for which direct sampling is difficult [ 29 – 31 ]. To obtain reliable posterior distributions, we run 50,000 iterations (5,000 burned), which proved to ensure the convergence of the MCMC algorithm.
The posterior distribution of θ B can be computed following the same steps. Once both posterior distributions, p ( θ A | x A ) and p ( θ B | x B ), are derived, we compute the distribution of the difference between the scaling exponents by subtracting the posteriors, i.e.
Then, by observing the 90% High Density Interval ( HDI 90) of p ( θ A − θ B ), we can draw a full probabilistic assessment of the similarity between the two distributions.
Acknowledgments
Funding for this work was provided by EU FET project MULTIPLEX nr. 317532, SIMPOL nr. 610704, DOLFINS nr. 640772, SOBIGDATA nr. 654024. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. We want to thank Geoff Hall and “Skepti Forum” for providing fundamental support in defining the atlas of conspiracy news sources in the US Facebook.
Data Availability
The entire data collection process has been carried out exclusively through the Facebook Graph API and the YouTube Data API. The collection methods are relayed in the Data Collection section.
Funding Statement
Funding for this work was provided by EU FET project MULTIPLEX nr. 317532, SIMPOL nr. 610704, DOLFINS nr. 640772, SOBIGDATA nr. 654024. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
References
Associated Data
Data Availability Statement
The entire data collection process has been carried out exclusively through the Facebook Graph API and the YouTube Data API. The collection methods are relayed in the Data Collection section.
Relations on this page carry a quotation for the claim. Where a relation was separately evidenced — and so tested against the vocabulary for its type — the row says so; 4 of 4 relations here carry that mark. The rest are the tagger’s reading of the sentence quoted beside them.
This record holds no relations between claims. A layer is empty when the tagger recorded nothing in it — which is not the same as the paper having nothing to record.
| facet | tag | field | conf. | whose words |
|---|---|---|---|---|
| object_of_study | how users polarize around science and conspiracy narratives across Facebook and YouTube | object_of_study | the paper's words | |
| research_question | whether content or platform-specific promotion algorithms better explain echo-chamber formation | research_question | the paper's words | |
| contribution | a cross-platform behavioral model that predicts later polarization from early commenting patterns | contribution | the paper's words | |
| claim_type | mechanism | claim_type | the paper's words | |
| phenomenon | users aggregate into homogeneous communities around conflicting narratives on both platforms | phenomenon | the paper's words | |
| method | network-analysis also in: network-analysis | method | the paper's words | |
| mechanisms | Disintermediated selection encourages users to choose belief-consistent information. | mechanisms[0] | 0.95 | the tagger's words |
| mechanisms | Personalized promotion can limit exposure to unsolicited viewpoints. | mechanisms[1] | 0.95 | the tagger's words |
| mechanisms | Confirmation-biased selection aggregates users into polarized communities. | mechanisms[2] | 0.95 | the tagger's words |
| mechanisms | Repeated engagement with one narrative creates preferential attachment to its community. | mechanisms[3] | 0.95 | the tagger's words |
| mechanisms | Conflicting content organizes homogeneous groups despite different promotion systems. | mechanisms[4] | 0.95 | the tagger's words |
| limitations | The design compares only science and conspiracy narratives. | limitations[0] | 0.95 | the tagger's words |
| limitations | The specific causal role of opaque promotion algorithms remained debated. | limitations[1] | 0.95 | the tagger's words |
| limitations | Source categories relied on page self-description and external monitoring groups. | limitations[2] | 0.95 | the tagger's words |
| limitations | YouTube sampling was limited to videos linked by the selected Facebook pages. | limitations[3] | 0.95 | the tagger's words |
| limitations | Private users and restricted content were unavailable to the analysis. | limitations[4] | 0.95 | the tagger's words |
| limitations | The observational comparison cannot independently manipulate content and algorithms. | limitations[5] | 0.95 | the tagger's words |
| controls | same videos on both platforms eliminates: content differences confounded with platform differences · outcome: held | controls[0] | 0.95 | the tagger's words |
| controls | Science-versus-Conspiracy contrast eliminates: mixing conflicting narrative categories · outcome: held | controls[1] | 0.95 | the tagger's words |
| controls | public API records only eliminates: inclusion of privacy-restricted user content · outcome: held | controls[2] | 0.95 | the tagger's words |
| controls | separate training and test sets eliminates: within-sample overfitting · outcome: held | controls[3] | 0.95 | the tagger's words |
| controls | Monte Carlo cross-validation eliminates: single random partition instability · outcome: held | controls[4] | 0.95 | the tagger's words |
| controls | cross-platform training and testing eliminates: platform-specific classification artifact · outcome: held | controls[5] | 0.95 | the tagger's words |
| concepts | confirmation bias also in: confirmation bias | concepts[0] | 0.95 | the paper's words |
| concepts | echo chamber also in: echo chamber | concepts[1] | 0.95 | the paper's words |
| concepts | polarization of users about this quote: This quote ends on 'as' because what it introduces is a displayed formula that the staged extraction does not carry as readable text (the text continues: "We then consider users with ρ u > 0.95 as users "). The quote is verbatim and the formula is in the paper; the quote was not extended, because extending it would put extraction debris inside a quotation. also in: polarization of users | concepts[2] | 0.95 | the paper's words |
| concepts | content promotion algorithm also in: content promotion algorithm | concepts[3] | 0.95 | the paper's words |
| concepts | scientific news also in: scientific news | concepts[4] | 0.95 | the paper's words |
| concepts | conspiracy-like news also in: conspiracy-like news | concepts[5] | 0.95 | the paper's words |
| concepts | commenting patterns also in: commenting patterns | concepts[6] | 0.95 | the paper's words |
| concepts | comments also in: comments | concepts[7] | 0.95 | the paper's words |
| concepts | multinomial logistic model also in: multinomial logistic model | concepts[8] | 0.95 | the paper's words |
| concepts | preferential attachment also in: preferential attachment | concepts[9] | 0.95 | the paper's words |
| concepts | collective debate also in: collective debate | concepts[10] | 0.95 | the paper's words |
| senses | bias - selective consumption of information adhering to prior beliefs | senses[0] | 0.95 | the tagger's words |
| senses | attention - aggregate likes, comments, shares, and views received by content | senses[1] | 0.95 | the tagger's words |
| senses | prediction - classification of future community polarization from early comments | senses[2] | 0.95 | the tagger's words |
| senses | reward - a like serving as positive feedback to online content | senses[3] | 0.95 | the tagger's words |
| senses | representation - a platform-visible science or conspiracy narrative category | senses[4] | 0.95 | the tagger's words |
| concept_edges | confirmation bias —causes→ echo chamber the relation itself is separately evidenced · strength of the relation: 0.95 · the relation stated: Belief-consistent selection leads users toward homogeneous communities.
Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but recent studies 'suggest' confirmation bias drives content selection toward polarized communities; hedged, and cited. also in: confirmation bias · echo chamber | concept_edges[0] | 0.95 | the tagger's words |
| concept_edges | content promotion algorithm —contrasted-with→ echo chamber the relation itself is separately evidenced · strength of the relation: 0.95 · the relation stated: Echo chambers emerge independently of the promotion algorithm.
Audit 2026-09-28: DOUBTFUL - read against its span and kept, but echo chambers form INDEPENDENTLY of the promotion algorithm; contrasted-with paraphrases independence. also in: content promotion algorithm · echo chamber | concept_edges[1] | 0.95 | the tagger's words |
| concept_edges | polarization of users —operationalised-as→ comments the relation itself is separately evidenced · strength of the relation: 0.95 · about this quote: This quote ends on 'as' because what it introduces is a displayed formula that the staged extraction does not carry as readable text (the text continues: "We then consider users with ρ u > 0.95 as users "). The quote is verbatim and the formula is in the paper; the quote was not extended, because extending it would put extraction debris inside a quotation. [about the separate quote for the relation] Update, 17 September 2026: This quote ends on 'as' because what it introduces is a displayed formula that the staged extraction does not carry as readable text (the text continues: "We then consider users with ρ u > 0.95 as users "). The quote is verbatim and the formula is in the paper; the quote was not extended, because extending it would put extraction debris inside a quotation. · the relation stated: Polarization is defined as a fraction of comments. also in: polarization of users · comments | concept_edges[2] | 0.95 | the tagger's words |
| concept_edges | scientific news —contrasted-with→ conspiracy-like news the relation itself is separately evidenced · strength of the relation: 0.95 · the relation stated: The study treats them as distinct conflicting narratives. also in: scientific news · conspiracy-like news | concept_edges[3] | 0.95 | the tagger's words |
| content_tags | Likes, comments, and shares showed strongly related patterns within each platform. | content_tags.findings[0] | 0.95 | the tagger's words |
| content_tags | Success measured as attention did not transfer reliably between platforms. | content_tags.findings[1] | 0.95 | the tagger's words |
| content_tags | Science and conspiracy videos received similar volumes of attention. | content_tags.findings[2] | 0.95 | the tagger's words |
| content_tags | Most users concentrated almost all comments on one of the two narratives. | content_tags.findings[3] | 0.95 | the tagger's words |
| content_tags | Content produced separated echo chambers independently of the promotion algorithm. | content_tags.findings[4] | 0.95 | the tagger's words |
| content_tags | Polarized users behaved similarly across conflicting narratives and platforms. | content_tags.findings[5] | 0.95 | the tagger's words |
| content_tags | Fifty comments supported classification accuracy above 0.80 for every class. | content_tags.findings[6] | 0.95 | the tagger's words |
| content_tags | A classifier trained on one platform generalized above 0.80 accuracy to the other. | content_tags.findings[7] | 0.95 | the tagger's words |
| content_tags | Conflicting narratives, rather than platform algorithm differences, explained homogeneous aggregation. | content_tags.findings[8] | 0.95 | the tagger's words |
| content_tags | selection of information that adheres to an existing belief system also in: confirmation bias | content_tags.definitions[0] | 0.95 | the tagger's words |
| content_tags | like-minded group sharing and polarizing around the same narrative also in: echo chamber | content_tags.definitions[1] | 0.95 | the tagger's words |
| content_tags | positive feedback to a post or video also in: like | content_tags.definitions[2] | 0.95 | the tagger's words |
| content_tags | an action intended to increase information visibility also in: share | content_tags.definitions[3] | 0.95 | the tagger's words |
| content_tags | an observable contribution to collective online debate also in: comment | content_tags.definitions[4] | 0.95 | the tagger's words |
| content_tags | fraction of a user's comments attached to conspiracy content about this quote: This quote ends on 'as' because what it introduces is a displayed formula that the staged extraction does not carry as readable text (the text continues: "We then consider users with ρ u > 0.95 as users "). The quote is verbatim and the formula is in the paper; the quote was not extended, because extending it would put extraction debris inside a quotation. also in: polarization of users | content_tags.definitions[5] | 0.95 | the tagger's words |
| content_tags | user with more than ninety-five percent of interactions in one narrative also in: polarized user | content_tags.definitions[6] | 0.95 | the tagger's words |
| content_tags | moment-based measure compared with a uniform-distribution benchmark also in: bimodality coefficient | content_tags.definitions[7] | 0.95 | the tagger's words |
| content_tags | classification model for a categorical outcome with more than two classes also in: multinomial logistic regression | content_tags.definitions[8] | 0.95 | the tagger's words |
| content_tags | 12 million users — cross-platform user sample described in the study | content_tags.quantities[0] | 0.95 | the tagger's words |
| content_tags | 413 public pages — US Facebook science and conspiracy sources | content_tags.quantities[1] | 0.95 | the tagger's words |
| content_tags | five years — Facebook observation period from 2010 through 2014 | content_tags.quantities[2] | 0.95 | the tagger's words |
| content_tags | about 17,000 videos — YouTube videos linked from selected Facebook posts | content_tags.quantities[3] | 0.95 | the tagger's words |
| content_tags | 400 users per class — samples used to model polarization trajectories | content_tags.quantities[4] | 0.95 | the tagger's words |
| content_tags | 50 comments — early-history compromise yielding accuracy above 0.80 | content_tags.quantities[5] | 0.95 | the tagger's words |
| content_tags | r = 0.987; simulated p < 0.01 — similarity of Science and Conspiracy action-correlation matrices | content_tags.statistics[0] | 0.95 | the tagger's words |
| content_tags | BC Facebook 0.964; BC YouTube 0.928 — bimodality of cross-platform polarization | content_tags.statistics[1] | 0.95 | the tagger's words |
| content_tags | 93.6% Facebook; 87.8% YouTube — users beyond the study's polarization thresholds | content_tags.statistics[2] | 0.95 | the tagger's words |
| content_tags | accuracy greater than 0.80 — cross-platform and within-platform classification performance | content_tags.statistics[3] | 0.95 | the tagger's words |
| content_tags | likes also in: likes | content_tags.variables[0] | 0.95 | the tagger's words |
| content_tags | comments also in: comments | content_tags.variables[1] | 0.95 | the tagger's words |
| content_tags | shares also in: shares | content_tags.variables[2] | 0.95 | the tagger's words |
| content_tags | polarization about this quote: This quote ends on 'as' because what it introduces is a displayed formula that the staged extraction does not carry as readable text (the text continues: "We then consider users with ρ u > 0.95 as users "). The quote is verbatim and the formula is in the paper; the quote was not extended, because extending it would put extraction debris inside a quotation. also in: polarization | content_tags.variables[3] | 0.95 | the tagger's words |
| content_tags | n comments also in: n comments | content_tags.variables[4] | 0.95 | the tagger's words |
Tags that could not be placed in the text:
What was deliberately not tagged, and why, is recorded per paper by the tagger. Where a paper has no such record, this page says so rather than implying that nothing was declined.