Dataset Card for Misviz

Dataset Summary

Misviz is a dataset of 2,604 real-world data visualizations collected from the web and manually annotated for misleading design practices. The dataset is introduced in the arXiv preprint Is this chart lying to me? Automating the detection of misleading visualizations.

Each visualization may contain up to three misleading design violations from a taxonomy of 12 misleaders. The dataset is designed to evaluate multimodal models for detecting misleading visualizations.

The dataset contains:

• The visualization image
• The original image URL
• A Wayback Machine fallback URL
• The chart type
• One or more misleaders
• Optional misleader bounding box annotations
• A predefined split field

---

Dataset Structure

The dataset consists of a single JSON file containing a list of instances.

Data Fields

Each entry contains the following fields:

• image
  The visualization image (loaded as an Image feature on Hugging Face).

• image_path
  Relative path to the image inside the repository.

• image_url
  Original URL from which the image was collected.

• wayback_image_url
  Archived URL used as fallback if the original is unavailable.

• chart_type
  A list containing one or more chart type labels

• misleader
  A list of misleading design violations affecting the visualization. Each visualization may contain up to three misleaders from a taxonomy of 12.

• split
  Split label (dev, val, or test).

• bbox
  Optional bounding box annotations for localizing misleaders.

Data Splits

The dataset contains predefined split labels:

• dev
• val
• test

On Hugging Face, the dataset is distributed as a single split for compatibility and simplicity with the main github repo. A specific split can be obtained as follows:

from datasets import load_dataset

ds = load_dataset("UKPLab/misviz", split="")
test_data = [x for x in ds if x["split"] == "test"]

Dataset Creation

Curation Rationale

Misleading visualizations can distort public understanding of data and contribute to misinformation.

Misviz was created to provide a benchmark for misleading visualization detection

The dataset includes original URLs and Wayback Machine URLs to ensure reproducibility.

Considerations for Using the Data

Social Impact of Dataset

This dataset supports research on detecting misleading visual content, which can help improve chart literacy, counter visual misinformation, and improve trust in data visualizations.

Known Limitations

• The dataset contains 2,604 visualizations, which is a moderate scale.

• Copyright of images remains with original creators.

• The dataset does not cover all types of misleaders covered in existing taxonomies

Licensing Information

The dataset annotations are released under a CC-BY-SA 4.0 license. We do not own copyright over the images. The dataset should be used only for academic research.

Citation Information

If you find this dataset useful, please cite the following paper

@article{tonglet2025misviz,
  title={Is this chart lying to me? Automating the detection of misleading visualizations},
  author={Tonglet, Jonathan and Zimny, Jan and Tuytelaars, Tinne and Gurevych, Iryna},
  journal={arXiv preprint arXiv:2508.21675},
  year={2025},
  url={https://arxiv.org/abs/2508.21675},
  doi={10.48550/arXiv.2508.21675}
}

Dataset Card Authors

Jonathan Tonglet

Dataset Card Contact

jonathan.tonglet@tu-darmstadt.de