DV-World: Benchmarking Data Visualization Agents in Real-World Scenarios

Jinxiang Meng, Shaoping Huang, Fangyu Lei, Jingyu Guo, Haoxiang Liu, Jiahao Su, Sihan Wang, Yao Wang, Enrui Wang, Ye Yang, Hongze Chai, Jinming Lyu, Anbang Yu, Huangjing Zhang, Yitong Zhang, Yiming Huang, Zeyao Ma, Shizhu He, Jun Zhao, Kang Liu
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:87882-87957, 2026.

Abstract

Real-world data visualization (DV) requires native environmental grounding, cross-platform evolution, and proactive intent alignment, yet existing benchmarks are often limited to code sandboxes, creation-only tasks, and fully specified intents. We introduce DV-World, a 260-task benchmark for evaluating DV agents across professional visualization lifecycles. DV-World covers three domains: DV-Sheet for native spreadsheet chart/dashboard creation and diagnostic repair; DV-Evolution for adapting reference visual artifacts to new data across programming paradigms; and DV-Interact for proactive intent alignment with a user simulator under ambiguous requirements. Its hybrid evaluation combines Table-value Alignment for numerical precision with rubric-based MLLM-as-a-Judge for semantic-visual assessment. Experiments show that state-of-the-art models achieve below 50% overall performance, exposing major gaps in real-world DV capabilities. DV-World provides a realistic testbed for developing DV agents suited to enterprise workflows. Data and code are available at dv-world-project.github.io.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-meng26d, title = {{DV}-World: Benchmarking Data Visualization Agents in Real-World Scenarios}, author = {Meng, Jinxiang and Huang, Shaoping and Lei, Fangyu and Guo, Jingyu and Liu, Haoxiang and Su, Jiahao and Wang, Sihan and Wang, Yao and Wang, Enrui and Yang, Ye and Chai, Hongze and Lyu, Jinming and Yu, Anbang and Zhang, Huangjing and Zhang, Yitong and Huang, Yiming and Ma, Zeyao and He, Shizhu and Zhao, Jun and Liu, Kang}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {87882--87957}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/meng26d/meng26d.pdf}, url = {https://proceedings.mlr.press/v306/meng26d.html}, abstract = {Real-world data visualization (DV) requires native environmental grounding, cross-platform evolution, and proactive intent alignment, yet existing benchmarks are often limited to code sandboxes, creation-only tasks, and fully specified intents. We introduce DV-World, a 260-task benchmark for evaluating DV agents across professional visualization lifecycles. DV-World covers three domains: DV-Sheet for native spreadsheet chart/dashboard creation and diagnostic repair; DV-Evolution for adapting reference visual artifacts to new data across programming paradigms; and DV-Interact for proactive intent alignment with a user simulator under ambiguous requirements. Its hybrid evaluation combines Table-value Alignment for numerical precision with rubric-based MLLM-as-a-Judge for semantic-visual assessment. Experiments show that state-of-the-art models achieve below 50% overall performance, exposing major gaps in real-world DV capabilities. DV-World provides a realistic testbed for developing DV agents suited to enterprise workflows. Data and code are available at dv-world-project.github.io.} }
Endnote
%0 Conference Paper %T DV-World: Benchmarking Data Visualization Agents in Real-World Scenarios %A Jinxiang Meng %A Shaoping Huang %A Fangyu Lei %A Jingyu Guo %A Haoxiang Liu %A Jiahao Su %A Sihan Wang %A Yao Wang %A Enrui Wang %A Ye Yang %A Hongze Chai %A Jinming Lyu %A Anbang Yu %A Huangjing Zhang %A Yitong Zhang %A Yiming Huang %A Zeyao Ma %A Shizhu He %A Jun Zhao %A Kang Liu %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-meng26d %I PMLR %P 87882--87957 %U https://proceedings.mlr.press/v306/meng26d.html %V 306 %X Real-world data visualization (DV) requires native environmental grounding, cross-platform evolution, and proactive intent alignment, yet existing benchmarks are often limited to code sandboxes, creation-only tasks, and fully specified intents. We introduce DV-World, a 260-task benchmark for evaluating DV agents across professional visualization lifecycles. DV-World covers three domains: DV-Sheet for native spreadsheet chart/dashboard creation and diagnostic repair; DV-Evolution for adapting reference visual artifacts to new data across programming paradigms; and DV-Interact for proactive intent alignment with a user simulator under ambiguous requirements. Its hybrid evaluation combines Table-value Alignment for numerical precision with rubric-based MLLM-as-a-Judge for semantic-visual assessment. Experiments show that state-of-the-art models achieve below 50% overall performance, exposing major gaps in real-world DV capabilities. DV-World provides a realistic testbed for developing DV agents suited to enterprise workflows. Data and code are available at dv-world-project.github.io.
APA
Meng, J., Huang, S., Lei, F., Guo, J., Liu, H., Su, J., Wang, S., Wang, Y., Wang, E., Yang, Y., Chai, H., Lyu, J., Yu, A., Zhang, H., Zhang, Y., Huang, Y., Ma, Z., He, S., Zhao, J. & Liu, K.. (2026). DV-World: Benchmarking Data Visualization Agents in Real-World Scenarios. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:87882-87957 Available from https://proceedings.mlr.press/v306/meng26d.html.

Related Material