Sign in to view source links and access this dataset
Description
AttackViz is a chart-image dataset introduced in the paper 'ChartAttack: Testing the Vulnerability of LLMs to Malicious Prompting in Chart Generation'. Each example contains a rendered chart image, metadata about the chart and question type, the expected gold answer, a binary label indicating whether the chart is correct or misleading, a misleading-visualization category, and serialized chart annotations. The dataset was created by jgermanmx and was last updated on Hugging Face in April 2026.
Use Cases
Testing LLM vulnerability to malicious prompting in chart generation based on the dataset's chart images and metadata.
Classifying charts as correct or misleading based on the provided binary labels.
Analyzing categories of misleading visualizations based on the provided misleading-visualization category field.
Training or evaluating computer vision models for chart understanding using the rendered chart images and serialized annotations.
Strengths
Includes a binary label for each chart indicating whether it is correct or misleading, enabling clear classification tasks.
Provides serialized chart annotations alongside rendered images, offering multimodal data for analysis.
Dataset is associated with a specific research paper ('ChartAttack'), providing a clear academic context and purpose.
Limitations
Row count is unknown, which may limit suitability assessment.
Column-level documentation is absent; field semantics must be inferred after download.
Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
Source
jgermanmx on Hugging Face.
Collection Method
Introduced in the research paper 'ChartAttack: Testing the Vulnerability of LLMs to Malicious Prompting in Chart Generation'; likely generated for that study.
Freshness
Last updated 2026-04-21 13:52:11; freshness should be verified.
License is unknown; users should verify terms of use before downloading.