A Survey of COVID-19 Misinformation: Datasets, Detection Techniques and Open Issues

A. R. Sana Ullah, Anupam Das, Anik Das, Muhammad Ashad Kabir, Kai Shu

Abstract

Misinformation during pandemic situations like COVID-19 is growing rapidly on social media and other platforms. This expeditious growth of misinformation creates adverse effects on the people living in the society. Researchers are trying their best to mitigate this problem using different approaches based on Machine Learning (ML), Deep Learning (DL), and Natural Language Processing (NLP). This survey aims to study different approaches of misinformation detection on COVID-19 in recent literature to help the researchers in this domain. More specifically, we review the different methods used for COVID-19 misinformation detection in their research with an overview of data pre-processing and feature extraction methods to get a better understanding of their work. We also summarize the existing datasets which can be used for further research. Finally, we discuss the limitations of the existing methods and highlight some potential future research directions along this dimension to combat the spreading of misinformation during a pandemic.

Introduction

The COVID-19 pandemic triggered a secondary crisis: an "infodemic" of rumors, fake news, and misleading health advice. This misinformation spreads faster than the virus itself, causing public panic and directly impacting health outcomes. While the research community responded with a flurry of detection systems, the field became fragmented. Researchers faced several challenges including data scarcity, language barriers, and methodological overlap. This paper addresses these gaps by providing a structured analysis of 47 core studies, focusing on data pre-processing, feature extraction, and classification methods used in recent literature.

Detection Methodologies

The survey categorizes detection techniques into three primary buckets based on their architectural complexity and approach to feature engineering.

Traditional Machine Learning (ML)

Algorithms like Support Vector Machines (SVM), Random Forest (RF), and XGBoost remain popular. These often rely on manual feature engineering, such as TF-IDF or Bag-of-Words (BoW), and psycholinguistic lexicons like LIWC.

Deep Learning (DL)

Deep Learning architectures are praised for their ability to capture complex patterns without manual feature extraction. The survey reviews fundamental Artificial Neural Networks (ANN) and more specialized structures.

Basic structure of an Artificial Neural Network (ANN) with input, hidden, and output layers.

Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs/LSTMs) are prominent in this category. CNNs, in particular, have shown exceptional performance in identifying local patterns within text sequences.

Architecture of a Convolutional Neural Network (CNN) showing convolution, pooling, and fully connected layers.

Transformers

Advanced models like BERT, RoBERTa, and ALBERT represent the current state-of-the-art in Natural Language Processing (NLP). Their bidirectional understanding of context allows them to handle the nuanced nature of health-related claims more effectively than standard DL models.

Datasets and Feature Extraction

Most research is centered on Twitter data, with labeled benchmarks like ReCOVery and Covid-HeRA being critical for model training. While English is dominant, emerging work is tackling multilingual datasets in Arabic, Chinese, and Indic languages. Feature extraction methods have evolved from simple word counts to deep contextualized word representations like ELMo and GloVe.

Key Results and Insights

  • High Accuracy: A CNN-based model achieved a staggering 99.99% accuracy on a specific dataset, while BERT-based models reached 99.20% in classifying rumors.
  • Explainability: The survey highlights models like dEFEND, which uses attention mechanisms to explain why a post was flagged, achieving 98% accuracy while providing necessary interpretability.
  • Ensemble Robustness: Combining traditional ML (SVM + Random Forest) yielded a robust 97.8% accuracy, proving that classical methods remain effective when combined correctly.
  • Pre-processing: Proper tokenization, stemming, and stopword removal remain foundational to high performance, a step often overlooked in favor of complex model architecture.

Open Issues and Future Directions

The authors identify several critical gaps for future research:

  • Multi-modality: Current research is largely text-centric; systems must evolve to detect misinformation in memes, images, and deepfake videos.
  • Supervision Burden: There is a dire need for unsupervised or semi-supervised learning techniques to adapt to new rumors in real-time without waiting for manual human annotation.
  • Class Imbalance: Most models are trained on balanced datasets, which may lead to high false-positive rates in real-world scenarios where fake news is a minority class.
  • Dynamic Nature: Misinformation evolves rapidly. Continuous, adaptive learning is necessary to ensure models remain relevant as pandemic trends shift.

Summary updated

Abstract

Misinformation during pandemic situations like COVID-19 is growing rapidly on social media and other platforms. This expeditious growth of misinformation creates adverse effects on the people living in the society. Researchers are trying their best to mitigate this problem using different approaches based on Machine Learning (ML), Deep Learning (DL), and Natural Language Processing (NLP). This survey aims to study different approaches of misinformation detection on COVID-19 in recent literature to help the researchers in this domain. More specifically, we review the different methods used for COVID-19 misinformation detection in their research with an overview of data pre-processing and feature extraction methods to get a better understanding of their work. We also summarize the existing datasets which can be used for further research. Finally, we discuss the limitations of the existing methods and highlight some potential future research directions along this dimension to combat the spreading of misinformation during a pandemic.
BibTeX
@inproceedings{ullah2022survey,
  author = {A. R. Sana Ullah and Anupam Das and Anik Das and Muhammad Ashad Kabir and Kai Shu},
  title = {{A Survey of COVID-19 Misinformation: Datasets, Detection Techniques and Open Issues}},
  booktitle = {Soc. Netw. Anal. Min},
  year = {2022},
  url = {https://openreview.net/forum?id=302I3Y9f45},
}
Figure