QCRI/COVID-19-disinformation

COVID-19 Infodemic Multilingual Dataset

This repository contains a multilingual dataset related to the COVID-19 infodemic, annotated with fine-grained labels. The dataset is curated to address questions of interest to journalists, fact-checkers, social media platforms, policymakers, and the general public. The dataset includes tweets in Arabic, Bulgarian, Dutch, and English, focusing on both binary (misinformation detection) and multiclass classification (different types of infodemic content).

Table of Contents:

• Dataset Overview
  • Languages and Splits
  • File Formats
• Annotations
• Dataset Examples
• Data Statistics
• License
• Citation

Dataset Overview

The dataset consists of tweets related to COVID-19, categorized under two tasks:

1. Binary Classification:
   Detecting whether a tweet contains misinformation.

2. Multiclass Classification:
   Classifying the tweet into specific infodemic categories such as conspiracy theories, harmful content, or false cures.

Languages and Splits

The dataset includes the following languages, each with train, development (dev), and test splits:

• Arabic
• Bulgarian
• Dutch
• English

In addition to individual language datasets, a multilang directory contains a multilingual dataset where tweets from all the above languages are combined in the binary and multiclass formats.

File Formats

The dataset is provided in TSV (Tab-Separated Values) format. Each file contains tweet IDs, labels for seven questions (Q1-Q7), and binary/multiclass annotations. The actual tweet text and associated metadata are not included for privacy reasons.

Directory Structure

• Readme.md: This file
• arabic/, bulgarian/, dutch/, english/: Directories containing language-specific datasets for both binary and multiclass classification.
• multilang/: A directory containing the multilingual version of the dataset.

Each language and the multilingual directory include three sets:

• train
• dev
• test

The *_binary_* files correspond to binary classification, while the *_multiclass_* files correspond to multiclass classification.

Annotations

The dataset contains labels for the following seven questions (Q1-Q7), each related to different aspects of the tweets:

1. Is the tweet understandable?

   • Labels: Yes, No, Not sure
   • This question evaluates whether the tweet's content is understandable.

2. Does the tweet contain false information?

   • Labels: Definitely no, Probably no, Not sure, Probably yes, Definitely yes
   • This question assesses the likelihood of false information in the tweet.

3. Will the tweet’s claim be of interest to the general public?

   • Labels: Definitely no, Probably no, Not sure, Probably yes, Definitely yes
   • Evaluates whether the tweet’s claim is relevant or interesting to the public.

4. Is the tweet harmful?

   • Labels: Definitely no, Probably no, Not sure, Probably yes, Definitely yes
   • Assesses if the tweet might cause harm to individuals, society, or businesses.

5. Should a professional fact-checker verify the claim?

   • Labels: No need, Too trivial, Not urgent, Very urgent, Not sure
   • Evaluates whether the tweet should be reviewed by fact-checkers.

6. Why might the tweet be harmful?

   • Labels: No harm, Panic, Hate speech, Rumor, Conspiracy, etc.
   • Categorizes the nature of potential harm the tweet might cause.

7. Should this tweet get the attention of a government entity?

   • Labels: Not interesting, Calls for action, Blames authorities, etc.
   • Determines if the tweet should be flagged for government attention.

Dataset Examples

An example from the dataset:

Tweet: "Please don’t take hydroxychloroquine (Plaquenil) plus Azithromycin for #COVID19 UNLESS your doctor prescribes it. Both drugs affect the QT interval of your heart and can lead to arrhythmias and sudden death, especially if you are taking other meds or have a heart condition."

Labels:

• Q1: Yes
• Q2: No, probably contains no false information
• Q3: Yes, definitely of interest
• Q4: No, probably not harmful
• Q5: Yes, very urgent
• Q6: No, not harmful
• Q7: Yes, calls for action

Data Statistics

• Arabic: 5,000 binary samples, 4,000 multiclass samples
• Bulgarian: 3,000 binary samples, 2,500 multiclass samples
• Dutch: 4,000 binary samples, 3,500 multiclass samples
• English: 6,000 binary samples, 5,000 multiclass samples
• Multilang: Combined data from all languages, provided in both binary and multiclass splits.

License

This dataset is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0).

Citation

If you use this dataset, please cite it as:

@inproceedings{alam-etal-2021-fighting-covid,
    title = "Fighting the {COVID}-19 Infodemic: Modeling the Perspective of Journalists, Fact-Checkers, Social Media Platforms, Policy Makers, and the Society",
    author = "Alam, Firoj  and
      Shaar, Shaden  and
      Dalvi, Fahim  and
      Sajjad, Hassan  and
      Nikolov, Alex  and
      Mubarak, Hamdy  and
      Da San Martino, Giovanni  and
      Abdelali, Ahmed  and
      Durrani, Nadir  and
      Darwish, Kareem  and
      Al-Homaid, Abdulaziz  and
      Zaghouani, Wajdi  and
      Caselli, Tommaso  and
      Danoe, Gijs  and
      Stolk, Friso  and
      Bruntink, Britt  and
      Nakov, Preslav",
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2021",
    month = nov,
    year = "2021",
    address = "Punta Cana, Dominican Republic",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.findings-emnlp.56",
    doi = "10.18653/v1/2021.findings-emnlp.56",
    pages = "611--649",

}

@inproceedings{alam2021fighting,
  title={Fighting the COVID-19 infodemic in social media: A holistic perspective and a call to arms},
  author={Alam, Firoj and Dalvi, Fahim and Shaar, Shaden and Durrani, Nadir and Mubarak, Hamdy and Nikolov, Alex and Da San Martino, Giovanni and Abdelali, Ahmed and Sajjad, Hassan and Darwish, Kareem and others},
  booktitle={Proceedings of the International AAAI Conference on Web and Social Media},
  volume={15},
  pages={913--922},
  year={2021}
}