Source-linked AI summary
Emotional Dynamics in the Age of Misinformation
Fabiana Zollo, Petra Kralj Novak, Michela Del Vicario, Alessandro Bessi, Igor Mozetic, Antonio Scala, Guido Caldarelli, Walter Quattrociocchi
TL;DR
The paper addresses how emotional dynamics differ between science and conspiracy news communities on social media. It analyzes Facebook comments using sentiment classification and finds more negative sentiment on conspiracy pages, with negativity increasing with user activity and discussion length. The study focuses on publicly available Facebook data and relies on aggregated sentiment to obtain stable values.
Problem
Unsubstantiated rumors and polarized, homophile online communities create a need to understand emotional dynamics around misinformation.
Method
The study analyzes four years of Italian Facebook science and conspiracy pages, applying supervised sentiment analysis to comments and examining polarized users and their interactions.
Results
Comments on conspiracy pages tend to be more negative than those on science pages, while negativity increases with polarized-user activity and discussion length.
Takeaways & Limitations
The findings connect emotional behavior within and across polarized science and conspiracy communities to misinformation-related content consumption and polarization.
Takeaways & Limitations
The dataset contains only publicly available Facebook data, excluding users whose privacy restrictions prevent access to their content.
Abstract
from arXiv · showhide
According to the World Economic Forum, the diffusion of unsubstantiated rumors on online social media is one of the main threats for our society. The disintermediated paradigm of content production and consumption on online social media might foster the formation of homophile communities (echo-chambers) around specific worldviews. Such a scenario has been shown to be a vivid environment for the diffusion of false claims, in particular with respect to conspiracy theories. Not rarely, viral phenomena trigger naive (and funny) social responses -- e.g., the recent case of Jade Helm 15 where a simple military exercise turned out to be perceived as the beginning of the civil war in the US. In this work, we address the emotional dynamics of collective debates around distinct kind of news -- i.e., science and conspiracy news -- and inside and across their respective polarized communities (science and conspiracy news). Our findings show that comments on conspiracy posts tend to be more negative than on science posts. However, the more the engagement of users, the more they tend to negative commenting (both on science and conspiracy). Finally, zooming in at the interaction among polarized communities, we find a general negative pattern. As the number of comments increases -- i.e., the discussion becomes longer -- the sentiment of the post is more and more negative.
Introduction
The paper examines how misinformation-related communities form and how emotional dynamics differ between science and conspiracy news. Using sentiment analysis of Facebook comments, it finds more negative sentiment on conspiracy pages and increasingly negative sentiment among more engaged polarized users and longer discussions.
- Introduction: Online platforms can foster selective exposure and homophile communities, reducing interaction with different narratives and producing polarized debates.
- Introduction: Unsubstantiated rumors can trigger naive responses and are identified by the World Economic Forum as a major societal threat.
- Introduction: The study analyzes four years of Italian Facebook data from scientific and conspiracy news sources distinguished by content verifiability.
- Introduction: Sentiment analysis uses annotated comments and a supervised SVM classifier assigning negative, neutral, or positive sentiment values.
- Introduction: The study links interactions between communities organized around shared narratives to emotional dynamics associated with misinformation spreading.
- Introduction: About 280k Italian Facebook users show more negative sentiment on conspiracy pages, with negativity increasing alongside polarized-user activity and discussion length.
Results and Discussion
Using supervised sentiment classification, the study finds consistently more negative emotional patterns around conspiracy content, polarized users, and longer discussions. Science posts become more positive with increased likes and shares, whereas comment-driven engagement is associated with increasing negativity across both categories.
- Sentiment classification: The classifier assigns comments to negative, neutral, or positive sentiment using a supervised machine-learning approach.The workflow includes manual annotation, classifier training, independent evaluation, and application to the full comment set.
- Sentiment on science and conspiracy posts: 70% of science-page comments are neutral or positive, compared with 51% on conspiracy pages; positive comments are 20% versus 10%.Science-page comments are twice as positive as conspiracy-page comments.
- Sentiment and virality: As comments increase, sentiment decreases for both science and conspiracy posts, while more likes and shares make science-post sentiment more positive.The relationships are estimated by regressing sentiment against the logarithm of comments, likes, or shares; randomization tests support the comment-related pattern.
- Sentiment and users activity: More active polarized users tend toward negative sentiment in both communities, with science users’ sentiment decreasing faster than conspiracy users’ sentiment.Polarized users are defined as leaving more than 95% of their likes on one content type; randomized tests support the activity-related association.
- Interaction across communities: Posts discussed by both polarized communities are 60% negative, and their sentiment becomes increasingly negative as the discussion lengthens.These cross-community posts show a higher concentration of negative sentiment than general posts.
Conclusions
The study finds generally more negative sentiment around conspiracy pages and increasingly negative sentiment as polarized users become more active or discussions lengthen.
- Conspiracy-page sentiment tends to be more negative than science-page sentiment.
Methods
The study analyzes public Italian Facebook pages and comments over four years, using manual annotation and supervised sentiment classification to compare science and conspiracy communities. It also identifies polarized users through category-specific liking behavior and examines sentiment across engagement thresholds.
- Data collection: Researchers analyzed public Facebook data exclusively through the Facebook Graph API, excluding users with privacy restrictions.The pages and contributing user content were publicly accessible under applicable privacy settings.
- Data collection: The dataset contains 73 public Italian Facebook pages collected from 2010 to 2014 and categorized as science or conspiracy sources.Pages were categorized using their contents and self-descriptions.
- Data annotation: A subset of 19,642 comments was manually annotated to train a sentiment classifier.Annotators labeled comments as negative, neutral, or positive, while performance was monitored using duplicated comments and agreement measures.
- Classification: Sentiment was treated as ordinal classification with three ordered classes and modeled using two linear-kernel SVM classifiers.The classifiers separated negative and positive comments from the remaining classes, assigning neutral when their predictions disagreed.
- Polarized users: Users were labeled polarized when more than 95% of their likes were directed to either science or conspiracy posts.The threshold-based labeling procedure also examined polarized-user counts and mean sentiment as the threshold changed.