Text2Vis is a benchmark dataset designed to evaluate the ability of large language models (LLMs) to generate accurate and high-quality data visualizations from natural language queries. It supports evaluation across multiple dimensions including answer correctness, chart quality, reasoning difficulty, and question answering over structured data.

🏆 Accepted to EMNLP 2025 (Main Conference)

You can access the dataset through the following links:

• 📦 Hugging Face Dataset
• 📁 Google Drive Download

• 📚 ArXiv Version

If you have any questions about this work, please contact **[Mizanur Rahman] mizanur.york@gmail.com.

If you use Text2Vis in your research, please cite:

@misc{rahman2025text2vischallengingdiversebenchmark,
      title={Text2Vis: A Challenging and Diverse Benchmark for Generating Multimodal Visualizations from Text}, 
      author={Mizanur Rahman and Md Tahmid Rahman Laskar and Shafiq Joty and Enamul Hoque},
      year={2025},
      eprint={2507.19969},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2507.19969}, 
}