During the COVID-19 pandemic, political messages often spread faster—and more powerfully—than scientific information. So why do people hold such radically different views about vaccines? And how closely are those views tied to political ideology? The study introduced today tackles these questions using a hybrid analytical approach that combines LDA, BERTopic, and Social Network Analysis. Analyzing large-scale Twitter data from Brazil, the researchers reveal how pro-vaccine, anti-vaccine, and anti-Sinovac groups differ in the topics they discuss, their political leanings, and their network structures.

Backgroud

Political Polarization Behind Vaccine Debates

Brazil traditionally maintains high vaccination rates, yet the early phase of the COVID-19 rollout showed a very different landscape. Mistrust in rapid vaccine development, anti-China sentiment, and intense conflicts between federal and state governments turned a scientific issue into a highly political one. The research team focused on a key question: How does political polarization shape people’s attitudes toward vaccination?

​

Core Idea

Viewing Vaccine Debates as Issues of Political Identity

The study begins with the premise that attitudes toward vaccines are not purely scientific judgments. They are deeply intertwined with political identity and group affiliation. • What topics people discuss • Their political leaning • Their network structures • The types of information sources they rely on In the topic modeling stage, a hybrid approach combining LDA and BERTopic captures both high-level themes and fine-grained opinion clusters.

​

Solution

The study’s main contribution is its integrated framework, combining multiple techniques to model public debate at multiple levels.

​

1. LDA: Mapping the High-Level Topic Landscape

LDA identifies the major issue clusters across groups (pro-vaccine, anti-vaccine, anti-Sinovac, and neutral).

These include debates on mandatory vaccination, distrust of Chinese vaccines, government criticism, scientific trust, and more.

2.BERTopic: Uncovering Fine-Grained Opinion Patterns

Each LDA topic is further analyzed using BERTopic, revealing how opinions diverge within the same theme. Because BERTopic clusters texts based on semantic similarity, it distinguishes nuances such as:

​

• Anti-vaccine users opposing mandates due to – perceived threats to personal freedom – distrust in government – emotional reactions like “We won’t be lab rats”

​

3. Political Orientation: Calculating Left–Right Leaning

Political leaning scores are derived from the political figures each user follows. This quantifies how tightly vaccine attitudes correlate with political identity.

​

4. Social Network Analysis (SNA): Community Structure and Echo Chambers

Follower networks reveal distinct political communities. Cluster cohesion, central influencers, and cross-group connections are analyzed to assess echo-chamber effects.

5. Information Source Analysis

By tracking shared links and mentions, the study shows how groups rely on entirely different media ecosystems (news outlets, YouTube, social media, or political figures).

​

Together, these steps form a full analytical chain: Topic → Opinion Pattern → Political Identity → Network Structure → Information Diffusion

​

Result

Vaccine Attitudes Are Strongly Tied to Political Identity

The patterns are remarkably consistent across groups:

​

• Pro-vaccine users → Left-leaning, trust in science, government criticism • Anti-vaccine users → Right-leaning, opposition to mandates, distrust in safety • Anti-Sinovac group → Strong far-right leaning, anti-China rhetoric, political conspiracy narratives

​

​

Network analyses show that all groups exhibit clear left/right divisions. Notably, the anti-vaccine group forms a dense, highly cohesive community where messages spread quickly and reinforce internal beliefs—classic echo-chamber behavior.

Right-leaning communities have tighter, faster diffusion networks, driven by political influencers and partisan media. Left-leaning communities are larger but more loosely connected, with journalists, science communicators, and activists acting as key intermediaries. These findings show how political polarization does more than shape opinions—it reshapes the network structure of public discourse itself.

Although focused on Brazil, this study offers insights relevant worldwide. Whether the topic is vaccines, climate change, or AI, scientific issues often become political when reinforced through polarized social networks. In a polarized era, who delivers the message can matter more than the scientific evidence itself. This research provides valuable implications for public health, online discourse analysis, topic modeling, and social network analysis.

​

NetMiner Team

www.netminer.com

netminer@cyram.com

References

[1] Ebeling, R., Sáenz, C. A. C., Nobre, J., & Becker, K. (2021). Analysis of the influence of political polarization in the vaccination stance: The Brazilian COVID-19 scenario [Preprint]. arXiv. https://arxiv.org/abs/2110.03382