Source-linked AI summary
Trend of Narratives in the Age of Misinformation
Alessandro Bessi, Fabiana Zollo, Michela Del Vicario, Antonio Scala, Guido Caldarelli, Walter Quattrociocchi
TL;DR
Social media’s direct producer-to-consumer path accompanies rumors and mistrust, making conspiracy narratives’ emergence and popularity important to study. This paper analyzes conspiracy-news pages on Italian Facebook using semi-automatic topic extraction; activity is similar across topics, but commenting persistence and topic mobility vary.
Problem
How conspiracy narratives emerge and become popular on social media merits study amid the production of rumors and mistrust.
Source: Introduction · p. 1Method
The study analyzes Italian Facebook conspiracy-news pages and uses semi-automatic topic extraction to characterize their content and consumption.
Source: Introduction · p. 2Results
Likes and comments are similar across topics, but geopolitics users are most persistent and diet users least; each same-category like raises the probability of moving topics by 12%.
Source: Introduction · p. 2, Conclusions · pp. 10–11Takeaways & Limitations
The findings provide insight into how conspiracy-like rumors are consumed and how misinformed narratives diffuse on online social media.
Source: Conclusions · p. 11Conspiratorial terms are included only when at least 90% of volunteers label them so, and the analysis considers only the frequentist approach.
Source: Topics extraction and validation · p. 3, Preliminaries and Definitions · p. 14
Abstract
from arXiv · showhide
Social media enabled a direct path from producer to consumer of contents changing the way users get informed, debate, and shape their worldviews. Such a {\em disintermediation} weakened consensus on social relevant issues in favor of rumors, mistrust, and fomented conspiracy thinking -- e.g., chem-trails inducing global warming, the link between vaccines and autism, or the New World Order conspiracy. In this work, we study through a thorough quantitative analysis how different conspiracy topics are consumed in the Italian Facebook. By means of a semi-automatic topic extraction strategy, we show that the most discussed contents semantically refer to four specific categories: {\em environment}, {\em diet}, {\em health}, and {\em geopolitics}. We find similar patterns by comparing users activity (likes and comments) on posts belonging to different semantic categories. However, if we focus on the lifetime -- i.e., the distance in time between the first and the last comment for each user -- we notice a remarkable difference within narratives -- e.g., users polarized on geopolitics are more persistent in commenting, whereas the less persistent are those focused on diet related topics. Finally, we model users mobility across various topics finding that the more a user is active, the more he is likely to join all topics. Once inside a conspiracy narrative users tend to embrace the overall corpus.
Introduction
The paper studies how conspiracy narratives are consumed on Italian Facebook amid social-media rumors and mistrust. It finds similar activity across topic categories, but differences in commenting persistence and topic mobility.
- Conspiracy narratives simplify causation and foster disengagement from mainstream society and recommended practices; adopted false beliefs are rarely corrected.Source: Introduction · p. 2
- Despite collective-intelligence rhetoric, socio-technical systems’ role in informed debate and their effects on public opinion remain unclear.Source: Introduction · p. 2
- The study analyzes Italian Facebook conspiracy-news pages, categorizing them by content and self-description.Source: Introduction · p. 2
- Likes and comments are similar across topics, but geopolitics users are most persistent in commenting and diet-focused users least persistent.Source: Introduction · p. 2
- The most discussed conspiracy contents refer to four topics: environment, diet, health, and geopolitics.Source: Introduction · p. 2
- Users can move independently between topics, and the probability of joining additional topics rises with user engagement.Source: Introduction · p. 2
Results and Discussion
The analysis characterizes conspiracy topics on Italian Facebook, their consumption patterns, and user mobility across topics. It identifies four categories and finds that engagement relates to broader topic participation, while user persistence differs by topic.
Topics extraction and validation
- The analysis labels 44,259 posts across environment, health, diet, and geopolitics using a semi-automatic topic strategy and a majority rule for mixed-topic posts.The classification uses volunteer tags with at least 90% concordance; ties are left unlabeled.Source: Topics extraction and validation · pp. 3–4
- Walktrap and Multilevel assign every term to the same community and match the manual classification with 100% accuracy; Fast greedy has 88.68% concordance.Source: Topics extraction and validation · p. 4, Figure 2 · p. 4
Content consumption
- Posts’ likes, comments, and shares are long-tailed and best fitted by discrete power laws; post survival functions do not differ significantly across topics.The Gehan-Wilcoxon test gives p = 0.091 for differences between topic survival functions.Source: Content consumption · pp. 4–6
Users activity
- Attention patterns of users polarized toward different conspiracy topics are comparable: despite some differences, all are well described by a power law.Geopolitics attracts 62.95% of polarized users, while diet attracts 5.94%. Diet receives as much attention as the other conspiracy topics, even though it is not a polarizing topic.Source: Users activity · pp. 6–7
- User persistence differs significantly across topics: geopolitics-focused users comment longest, while diet-focused users are least persistent.The Gehan-Wilcoxon test finds a difference among the four survival functions (all p-values < 10^-6).Source: Users activity · pp. 7–8
Modeling user mobility
- More engaged users are more likely to consider more topics, and with increasing engagement they tend to embrace the overall corpus without a specific pattern.The proportional-odds model associates an increase in likes with 12% increased odds of considering a higher number of topics.Source: Modeling user mobility · pp. 8–10
Conclusions
The paper examines how conspiracy narratives are consumed on Italian Facebook, finding similar activity across topics but differences in users’ commenting persistence and movement between topics.
Conclusions
- Italian Facebook conspiracy narratives provide a setting to study the emergence, production, and popularity of rumors, mistrust, and conspiracy-like accounts of reality.The paper describes these narratives as aimed at explaining and oversimplifying reality and its phenomena.Source: Conclusions · p. 10
- Likes and comments are similar across categories, but geopolitics-focused users comment most persistently and diet-focused users least persistently.Source: Conclusions · p. 10
- Users can move independently between topics, and each new same-category like raises the probability of moving to another topic by 12%.Source: Conclusions · pp. 10–11
Methods
The study analyzes posts and user interactions from 39 Italian Facebook conspiracy pages collected from 2010 to 2014. Its methods combine term co-occurrence networks, network analysis, and statistical models of commenting and topic activity.
Data Collection
- From 2010 to 2014, the dataset covers 39 Italian Facebook conspiracy pages, with around 9M likes and 1M comments from about 1.2M and 280K users, respectively.The dataset is the exhaustive set of conspiracy pages used for this analysis.Source: Data Collection · p. 12
Preliminaries and Definitions
- Posts and conspiracy terms form a bipartite graph, whose edges mark term presence in posts; its matrix supports a weighted term co-occurrence network.The co-occurrence matrix is defined as CA = MM^T.Source: Preliminaries and Definitions · p. 12
- The disparity filter extracts a weighted-network backbone while preserving its multi-scale nature, identifying significant heterogeneous links against a uniform-weight null model.A lower significance level filters out additional links.Source: Preliminaries and Definitions · p. 13
- Three community-detection algorithms—Walktrap, Multilevel, and Fast greedy—are applied to the term-network backbone to validate the manual classification.They use random-walk similarity, multilevel modularity optimization, and hierarchical agglomeration, respectively.Source: Preliminaries and Definitions · p. 13
- The Kaplan–Meier estimator measures the probability that an event, such as continued commenting, lasts beyond a given time.At each time step, estimated survival probability is the number still commenting minus those who stop, divided by the number still commenting.Source: Preliminaries and Definitions · pp. 13–14
- The proportional odds model handles ordered categorical responses by allowing category-specific intercepts while sharing slopes across response cut-points.It combines information across possible cut-point logistic regressions rather than relying on one arbitrary cut-point.Source: Preliminaries and Definitions · p. 14