Source-linked AI summary
Echo Chambers: Emotional Contagion and Group Polarization on Facebook
Michela Del Vicario, Gianna Vivaldo, Alessandro Bessi, Fabiana Zollo, Antonio Scala, Guido Caldarelli, Walter Quattrociocchi
TL;DR
Limited evidence exists on how polarized Facebook communities evolve structurally and how user engagement shapes them. This paper compares Italian science and conspiracy communities using growth models and sentiment measures, finding similar size dynamics and greater negativity among more involved users.
Problem
Little is known about the structural evolution of science and conspiracy echo chambers and the role of user engagement in shaping them.
Method
The study compares Italian science and conspiracy Facebook communities using daily growth-model fits and measures of users’ sentiment, activity, and involvement.
Results
Both communities show rapid initial growth followed by gradual expansion toward a threshold, while greater involvement corresponds to more negative sentiment and faster shifts among more active users.
Takeaways & Limitations
Science and conspiracy echo chambers display similar growth patterns, and engagement is associated with increasingly negative emotional behavior in both communities.
Takeaways & Limitations
The analysis uses publicly available Facebook data and excludes users whose privacy restrictions prevent their content from being accessed.
Abstract
from arXiv · showhide
Recent findings showed that users on Facebook tend to select information that adhere to their system of beliefs and to form polarized groups -- i.e., echo chambers. Such a tendency dominates information cascades and might affect public debates on social relevant issues. In this work we explore the structural evolution of communities of interest by accounting for users emotions and engagement. Focusing on the Facebook pages reporting on scientific and conspiracy content, we characterize the evolution of the size of the two communities by fitting daily resolution data with three growth models -- i.e. the Gompertz model, the Logistic model, and the Log-logistic model. Then, we explore the interplay between emotional state and engagement of users in the group dynamics. Our findings show that communities' emotional behavior is affected by the users' involvement inside the echo chamber. Indeed, to an higher involvement corresponds a more negative approach. Moreover, we observe that, on average, more active users show a faster shift towards the negativity than less active ones.
Introduction
The introduction frames digital misinformation and polarized echo chambers as societal risks, then presents a comparative study of Italian Facebook science and conspiracy communities. It examines community evolution alongside user engagement and emotional dynamics, finding that greater involvement and activity correspond to faster movement toward negativity.
- Problem: Digital misinformation has expanded with new communication technologies and remains a major societal threat, while existing mitigation strategies appear ineffective.The introduction also links misinformation to consequential public decisions.
- Background: Disintermediated social media selection promotes confirmation bias and like-minded groups that polarize opinions, while dissenting information is often ignored or intensifies polarization.Confirmatory information may be accepted even when deliberately false.
- Contribution: The study comparatively analyzes polarized science and conspiracy communities on Italian Facebook, examining their temporal size evolution and community behavior.The analysis accounts for user engagement and emotional dynamics.
- Results: Greater involvement in either echo chamber corresponds to a more negative emotional state, and more active users shift toward negativity faster than less active users.The involvement-related emotional pattern appears in both user categories.
Results and Discussion
Science and conspiracy communities show similar growth dynamics and user behavior, with more than 99% of user flow entering communities. Greater engagement is associated with increasingly negative sentiment and faster shifts toward negativity, especially among highly active science users.
- Community growth: Over 835 daily observations from January 2010–April 2012, science and conspiracy communities exhibit similar global growth with quantitative differences by activity threshold.Differences arise between C1/C2 and C5, and between C1/C2 and the corresponding Si communities.
- Community growth: More than 99% of user flow consists of entries, and communities evolve after initial spike-like growth at a nearly constant rate.The same pattern holds for both science and conspiracy communities, regardless of content.
- Community growth: All tested growth models approximate community sizes well, but their correlations with trends are too similar to identify a significantly preferred fit.This agrees with the absence of a meaningful growth difference between science and conspiracy communities.
- Community growth: The first two temporal principal components capture 96.16% of S5 variance and 95.44% of C5 variance, corresponding to each series’ trend.No significant cycles remain after trend removal, indicating that trends determine the records’ time evolution.
- Emotional dynamics: As comment counts increase, mean user sentiment becomes more negative, and greater engagement increases users’ tendency to express negative emotion.This pattern holds for both science and conspiracy users.
- Emotional dynamics: More active users shift toward negativity faster, with the rate higher above 100 comments and higher for science users than conspiracy users.Cumulative activity also produces a larger shift toward negative comments in both communities.
Conclusions
The study characterizes polarized Facebook communities by modeling their structural evolution alongside users’ activity and sentiment. Both communities show similar growth dynamics, while greater involvement is associated with more negative emotional behavior and faster shifts toward negativity among more active users.
- Conclusions: Facebook’s environment supports polarized communities whose activity is limited to a single type of content.The study examines users’ behavior and community evolution using both activity and expressed sentiment.
- Conclusions: Both communities show similar growth: rapid, approximately exponential expansion is followed by gradual growth until a thresholding value is reached.Daily data were fitted with the Gompertz, Logistic, and Log-logistic growth models.
- Conclusions: Higher involvement corresponds to a more negative emotional approach in both users and communities.On average, more active users shift toward negativity faster than less active users; this rate is higher above 100 comments and among science users versus conspiracy users.
Methods
The study uses publicly available Facebook data to compare science-information and conspiracy-theory communities. It fits community growth and analyzes dynamics using nonlinear least squares, goodness-of-fit testing, singular-spectrum analysis, and supervised sentiment classification.
- Data collection: Researchers collected publicly available data through the Facebook Graph API, excluding users with privacy restrictions.The analyzed pages and contributing user content were public unless privacy settings restricted access.
- Community identification: The analysis distinguishes conspiracy-theory pages from science-information pages using support from active misinformation-debunking Facebook groups.Conspiracy pages promote content neglected by mainstream media, whereas science pages diffuse checkable scientific news and research advances.
- Growth modeling: Community sizes are fitted with Gompertz, Logistic, and Log-Logistic growth models, including the L3 and L5 Logistic variants.L3 uses parameters (b,d,f), while L5 corresponds to the full Logistic model.
- Model estimation: Model parameters are estimated by Nonlinear Least Squares, and fit goodness is tested with the Kolmogorov-Smirnov Test.The fitting procedure minimizes the sum of squared residual errors.
- Signal analysis: The study applies Singular-spectrum analysis and Monte-Carlo SSA to decompose signals and distinguish meaningful dynamics from background noise.SSA requires no assumed generating model, stationarity, or ergodicity conditions; MCSSA addresses signal-to-noise separation.
- Sentiment classification: Comments receive negative (-1), neutral (0), or positive (+1) sentiment labels through supervised machine-learning classification trained on manually annotated texts.The training sample contains 20K randomly selected comments annotated by 22 native Italian speakers.