Papers by Andrew Reeson
Do Text-to-Vis Benchmarks Test Real Use of Visualisations? (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing benchmarks for visualisations are limited and do not reflect real-world use. |
| Approach: | They analysed visualisation code from Python, R, Javascript, and Vega to find similarities and differences between real-world and benchmark datasets. |
| Outcome: | The results show that benchmark datasets do not test the same distribution of chart types, attributes, and actions as real-world examples. |