cs.CYMar 4, 2026

Who Shapes Brazil's Vaccine Debate? Semi-Supervised Modeling of Stance and Polarization in YouTube's Media Ecosystem

Authors: Geovana S. de Oliveira, Ana P. C. Silva, Fabricio Murai, Carlos H. G. Ferreira

Organizations: Universidade Federal de Ouro Preto, João Monlevade, Brazil · Universidade Federal de Minas Gerais, Belo Horizonte, Brazil · Worcester Polytechnic Institute, Worcester, United States · Universidade Federal de Ouro Preto, Ouro Preto, Brazil

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Vaccination remains a cornerstone of global public health, yet the COVID-19 pandemic exposed how online misinformation, political polarization, and declining institutional trust can undermine immunization efforts. Most of the prior computational studies that analyzed vaccine discourse on social platforms focus on English-language data, specific vaccines, or short time windows, impairing our understanding of long-term dynamics in high-impact, non-English contexts like Brazil, home to one of the world's most comprehensive immunization systems. We here present the largest longitudinal study of Brazil's vaccine discourse on YouTube, leveraging a semi-supervised stance detection framework that combines self-labeling and self-training to classify nearly 1.4 million comments. By integrating stance with temporal patterns, engagement metrics, and channel taxonomy (legacy media, science communicators, digital-native outlets), we map how pro- and anti-vaccine narratives evolve and circulate within a hybrid media ecosystem. Our results show that semi-supervised learning substantially improves stance classification robustness, enabling fine-grained tracking of public attitudes across Brazil's full immunization schedule. Polarization spikes during epidemiological crises, especially COVID-19, but becomes fragmented across vaccines and interaction patterns in the post-pandemic period. Notably, science communication and digital-native channels emerge as the primary loci of both supportive and oppositional engagement, revealing structural vulnerabilities in contemporary health communication. Thus, our work advances computational methods for large-scale stance modeling while offering actionable evidence for public health agencies, platform governance, and online information ecosystems.

Figures & tables

Figure 1. Comparison between the final baseline and low-entropy models with 95% confidence intervals. A horizontal set of four bar charts comparing a Baseline model in blue and a Low Entropy model in gold. The first three charts show Precision, Recall, and F1-Score for three categories: Against, Favorable, and Inconclusive. In almost all categories, the gold bars (Low Entropy) are higher than the blue bars (Baseline). Specifically, for the Favorable class, the Baseline has a significant performance drop with large error bars, while the Low Entropy model maintains high performance and stability. The fourth chart on the right shows Overall Accuracy, where the Low Entropy model (gold) reaches approximately 0.9, clearly outperforming the Baseline which is at 0.85. All charts include black vertical error bars indicating 95 percent confidence intervals.
Stance# Comments# Likes# Replies# Users
Favorable152,940598,78066,95991,633
Against204,179838,97779,137115,037
Inconclusive1,039,5383,936,064475,114487,778
Table 1. Distribution of comments and engagement by stance.
Figure 2. Distribution of vaccine-specific mentions over the years (z-score normalized per vaccine). Two side-by-side heatmaps showing vaccine mentions from 2018 to 2024. The left heatmap (a) is in green scales and represents Pro-vaccine mentions. The right heatmap (b) is in red scales and represents Anti-vaccine mentions. The vertical axis lists 17 different vaccines, including BCG, COVID-19, HPV, and MMR. The horizontal axis represents the years. The color intensity indicates the Z-Score normalized volume, ranging from negative 1.0 (lightest) to positive 1.0 (darkest). In both charts, there is a clear concentration of darker cells (higher volume) starting in 2020 and 2021, especially for the COVID-19 vaccine, which shows peak intensity across both pro and anti sentiments during the pandemic years. Some vaccines like Yellow Fever show a specific peak in 2018, while others like HPV and MMR show more distributed mentions over the years.
Figure 3. Conditional reply probabilities by stance in pre, during, and post-pandemic periods. A set of three side-by-side bar charts titled Reply Probability. Each chart shows the probability of a reply being Against (red) or Favorable (green) based on the Parent Comment Class (Against or Favorable) on the horizontal axis. In the first chart, there is a clear echo effect: Against parents have a 0.59 probability of Against replies, while Favorable parents have a 0.69 probability of Favorable replies. The second chart shows a more balanced distribution, with probabilities hovering around 0.43 to 0.57 for both classes. The third chart shows a different trend where Against parent comments receive Against replies with 0.67 probability, but Favorable parents receive an equal split of 0.50 for both Against and Favorable replies. The vertical axis ranges from 0.0 to 1.0, and exact probability values are printed above each bar.
PeriodTotal of# PolarizedPolarized
CommentsCommentsProportion
Pre-pandemic74,17410,10413.62%
During Pandemic1,219,820317,40626.02%
Post-pandemic102,66329,60928.84%
Table 2. Distribution of polarized comments across pre, during, and post-pandemic periods.
Table 3. Cross-ranking Analysis: ANJ certification indicated (✓ Yes, ✗ No, – N/A). Ch. Type key: SC=Science Communicator, LNM=Legacy News Media, DC=Digital-native Commentary. Channels are sorted by total comments.
Ranked by counts of anti-vaccine comments
RankVideo Title (Translated)Video Link# Anti Comments# Views# Likes
1Covid-19: what might happen to those who reject the vaccine?Watch3,796509,38323,922
2Three diseases that are after-effects of Covid and nobody talks about!Watch3,1092,528,525138,783
3Covid vaccine: does AstraZeneca cause thrombosis?Watch3,0272,117,694100,382
4Health professionals dismiss the link between a teenager’s death and vaccinationWatch2,667285,22815,687
5The search for people who may never catch CovidWatch2,3083,120,004177,076
Table 4. Top 15 videos ranked by different counts. (Top) Largest number of anti-vaccine comments; (Middle) Largest number of pro-vaccine comments; (Bottom) Largest number of polarized comments, i.e., opposing + favorable. Titles were translated to English.
More in this topic

Social Media Analysis

Social media analysis studies online posts, media, users, and interactions to understand communication, communities, and public behavior.

See all 214 papers

Explore similar work

CardsList
Aug 18, 2026

A Decade of Climate Polarization on Brazilian YouTube using Language Models

Online platforms have become arenas for the public contestation of climate change, shaping how scientific knowledge, denial, and uncertainty are expressed and disputed. Yet longitudinal evidence remains limited for YouTube, especially for Portuguese-language discourse. Addressing this gap, we characterize how climate stances are expressed and contested over time in a large corpus of Portuguese-language YouTube comments retrieved through Brazil-oriented climate-related searches. To support this analysis in a noisy, imbalanced, and low-resource setting, we collect more than 240,000 comments posted between 2014 and 2024 and formulate stance detection as a three-way classification task (Believer, Denier, and Inconclusive). We operationalize stance attribution through a scalable self-training pipeline based on Llama 3.1, using Low-Rank Adaptation (LoRA) and hybrid instance selection to expand the training set with high-confidence pseudo-labeled examples while preserving class diversity. This approach improves coverage and class balance for minority and rhetorically complex classes, enabling large-scale stance attribution without extensive manual annotation. Our results show that polarisation is marked by interactional asymmetries: denialist comments are less prevalent, but they are associated with a comparatively higher share of cross-stance contestation, while pro-consensus discourse is more strongly reinforced within stance-homogeneous threads.

Daniel Morais, Diego H. M. Magalhaes, Gabriel H. Silva +4

Department of Computing and Systems,Universidade Federal de Ouro Preto,Brazil · ISTI, National Research Council (CNR),Italy · Department of Computer Science, University of Pisa,Italy +1

Stance DetectionPolitical Discourse Analysis
AskPDF
May 4, 2026

Mapping Discourse Reframing: A Multi-Layer Network Approach to Italian HPV Vaccine Discourse on X (2010-2024)

Understanding how online narratives travel through coalitions is critical for identifying information disorder, yet computational analyses often rely on conservative network constructions that erase initially sparse but salient signals. This paper proposes a novel multi-layer framework that captures low-frequency signals of emerging information disorder allowing for locating where online discourse is reframed and amplified over time. The use case is 14 years of Italian discourse on X regarding the Human Papillomavirus (HPV) vaccine across three pivotal epochs (2010-2024). Utilizing hashtag co-occurrence networks, we introduce a dual-layer approach. We first identify robust core discourse coalitions through conservative community detection, revealing a stable prevention-oriented backbone contrasted with increasingly separable skepticism coalitions. We then introduce a coverage layer and project fringe hashtags into core coalitions based on weighted connectivity. Using a manually labelled set of skeptical and conspiratorial seed tweets, we demonstrate that this core-coverage projection significantly improves the recovery of long-tail, problematic hashtags while preserving an interpretable coalition structure. Our findings characterize the structural maturation of polarized narratives and provide a methodology for mapping how discourse is reframed and amplified by information disorder over time.

Lorella Viola

Vrije Universiteit Amsterdam · De Boelelaan 1105, 1081 HV Amsterdam

Social Media Analysis
AskPDF
Mar 25, 2026

WhatsApp Vaccine Discourse (WhaVax): An Expert-Annotated Dataset and Benchmark for Health Misinformation Detection

We introduce WhaVax, a new expert-annotated dataset of vaccine-related WhatsApp messages collected from large Brazilian public groups spanning multiple pandemic years. The dataset was constructed through a rigorous, carefully designed pipeline that integrates keyword-based data collection, semantic deduplication to remove near-duplicate content, and a multi-stage annotation protocol conducted by medical specialists. This process produced a high-quality gold-standard corpus, characterized by substantial inter-annotator agreement and strong reliability for downstream analysis. Additionally, we provide a detailed characterization of WhatsApp misinformation, revealing distinctive linguistic, structural, lexical, temporal, and group-level patterns, as well as a meaningful layer of ambiguous cases that reflect the complexity of health discourse in private messaging. We also benchmark classical models, fine-tuned Small Language Models, and zero- or few-shot Large Language Models under realistic data-scarcity constraints, demonstrating that strong embeddings and LLM approaches perform competitively, while domain alignment and data availability remain critical factors. This study provides a rare, high-quality resource to support misinformation research and computational modeling in encrypted communication environments.

Jônatas H. dos Santos, Julio C. S. Reis, Philipe Melo +9

Universidade Federal de Minas Gerais (UFMG), Brasil · Universidade Federal de Viçosa (UFV), Brasil

Health Misinformation DetectionFake News DetectionAutomated Fact-Checking
AskPDF