A Second Pandemic? Analysis of Fake News About COVID-19 Vaccines in Qatar

Preslav Nakov, Firoj Alam, Shaden Shaar, Giovanni Da San Martino, Yifan Zhang

Abstract

While COVID-19 vaccines are finally becoming widely available, a second pandemic that revolves around the circulation of anti-vaxxer fake news may hinder efforts to recover from the first one. With this in mind, we performed an extensive analysis of Arabic and English tweets about COVID-19 vaccines, with focus on messages originating from Qatar. We found that Arabic tweets contain a lot of false information and rumors, while English tweets are mostly factual. However, English tweets are much more propagandistic than Arabic ones. In terms of propaganda techniques, about half of the Arabic tweets express doubt, and 1/5 use loaded language, while English tweets are abundant in loaded language, exaggeration, fear, name-calling, doubt, and flag-waving. Finally, in terms of framing, Arabic tweets adopt a health and safety perspective, while in English economic concerns dominate.

Introduction: The Rise of the Infodemic

As the world transitioned from managing the initial COVID-19 crisis to rolling out global vaccination programs, a parallel crisis emerged: the "Infodemic." This surge of disinformation and anti-vaxxer sentiment posed a significant threat to public health initiatives. While much of the existing research on social media misinformation has focused on Western contexts, this study examines Qatar—a unique, multilingual environment characterized by a large expatriate population living alongside local citizens.

The research aims to move beyond simple fact-checking to understand the construction of influence. Specifically, the study analyzes three dimensions of social media discourse:

  1. Factuality: The veracity of claims.
  2. Propaganda: The use of psychological triggers to manipulate opinion.
  3. Framing: The thematic "lens" (e.g., health vs. economy) used to present information.

Methodology

The researchers developed a multi-stage analysis pipeline to process and categorize content in both Arabic and English:

  • Disinformation Analysis: A Support Vector Machine (SVM) classifier was trained on a 7-question schema to identify verifiable claims, harmful content, and messages requiring fact-checking.
  • Propaganda Detection: Two specialized systems were utilized: Proppy, a Maximum Entropy classifier for stylistic "propagandistic" features, and Prta, which identifies 18 specific techniques such as "Loaded Language" and "Flag-waving."
  • Framing Analysis: The Tanbih system, based on a fine-tuned BERT model, categorized tweets into 15 media frames (e.g., Economy, Morality, Health and Safety).
  • Cross-Language Strategy: To leverage English-centric propaganda and framing tools, Arabic tweets were translated using the Google Translation API before analysis.

Data Collection

The study utilized a custom corpus of tweets originating from Qatar:

  • Arabic Corpus: 606 tweets collected between February 2020 and March 2021.
  • English Corpus: 2,589 tweets, divided into pre-vaccine and post-vaccine availability periods.
  • Training Data: Classifiers were trained on an established dataset containing 504 English and 218 Arabic tweets annotated for factuality and harm.

Key Findings: Linguistic and Cultural Divergence

The analysis revealed a striking divergence in how the pandemic and vaccines were discussed across different languages within the same geographic region.

Rumors vs. Manipulative Rhetoric

Arabic tweets were found to contain significantly more "fake news" and rumors. Conversely, English tweets were generally more factual but significantly more manipulative. Approximately 33% of English tweets were flagged as propagandistic, a rate much higher than that found in the Arabic sample.

Propaganda Techniques

The "toolkit" of manipulation varied by language. In Arabic discourse, "Doubt" was the dominant technique, appearing in 50% of propagandistic cases. English tweets utilized a broader array of techniques, including "Loaded Language," "Exaggeration," and "Fear." Notably, as vaccine rollouts began in late 2020, the use of "Loaded Language" in English tweets surged to 50%.

Framing and Demographics

The framing of the pandemic appeared to reflect the demographics of Qatar's population. Arabic tweets, likely originating from local citizens, primarily used a Health and Safety frame, reflecting community-centric concerns. English tweets, likely originating from the large expatriate workforce, were dominated by Economic concerns, reflecting anxieties regarding job security and the financial impact of lockdowns.

Discussion and Limitations

The study highlights that combating misinformation requires understanding the psychological and economic lenses of different communities. However, several limitations remain:

  • Platform Scope: The study is limited to Twitter, potentially missing the rapid spread of disinformation on "dark social" channels like WhatsApp.
  • Translation Nuance: Reliance on machine translation for Arabic propaganda detection may overlook subtle cultural idioms.
  • Global Applicability: While providing a snapshot of Qatar, the methodology needs expansion to other languages and platforms for a global understanding of the Infodemic.

Ultimately, this research underscores that the "second pandemic" of misinformation is as complex and multifaceted as the virus itself, requiring nuanced, language-specific strategies for mitigation.

Summary updated

Abstract

While COVID-19 vaccines are finally becoming widely available, a second pandemic that revolves around the circulation of anti-vaxxer fake news may hinder efforts to recover from the first one. With this in mind, we performed an extensive analysis of Arabic and English tweets about COVID-19 vaccines, with focus on messages originating from Qatar. We found that Arabic tweets contain a lot of false information and rumors, while English tweets are mostly factual. However, English tweets are much more propagandistic than Arabic ones. In terms of propaganda techniques, about half of the Arabic tweets express doubt, and 1/5 use loaded language, while English tweets are abundant in loaded language, exaggeration, fear, name-calling, doubt, and flag-waving. Finally, in terms of framing, Arabic tweets adopt a health and safety perspective, while in English economic concerns dominate.
BibTeX
@inproceedings{nakov2021second,
  author = {Preslav Nakov and Firoj Alam and Shaden Shaar and Giovanni Da San Martino and Yifan Zhang},
  title = {{A Second Pandemic? Analysis of Fake News About COVID-19 Vaccines in Qatar}},
  booktitle = {RANLP},
  year = {2021},
  url = {https://openreview.net/forum?id=amABgm5Cfq0},
}
Figure