IEEE VIS 2024 Content: VisEval: A Benchmark for Data Visualization in the Era of Large Language Models

Best Paper Award

VisEval: A Benchmark for Data Visualization in the Era of Large Language Models

Nan Chen - Microsoft Research, Shanghai, China

Yuge Zhang - Microsoft Research, Shanghai, China

Jiahang Xu - Microsoft Research, Shanghai, China

Kan Ren - ShanghaiTech University, Shanghai, China

Yuqing Yang - Microsoft Research, Shanghai, China

Room: Bayshore I + II + III

2024-10-15T16:40:00Z GMT-0600 Change your timezone on the schedule page
2024-10-15T16:40:00Z
Exemplar figure, described by caption below
Examples of visualization issues detected by VisEval: Llama (CodeLlama-7B) produces code that cannot be executed, while Gemini (Gemini-Pro) incorrectly maps the "sum of Tonnage" to the y-axis instead of "count" and lacks a legend for the "Cargo ship" color. GPT-3.5 fails to sort as specified and places the legend outside the canvas. Although GPT-4 almost meets the requirements, it still encounters overflow issues that impact readability.
Fast forward
Keywords

Visualization evaluation, automatic visualization, large language models, benchmark

Abstract

Translating natural language to visualization (NL2VIS) has shown great promise for visual data analysis, but it remains a challenging task that requires multiple low-level implementations, such as natural language processing and visualization design. Recent advancements in pre-trained large language models (LLMs) are opening new avenues for generating visualizations from natural language. However, the lack of a comprehensive and reliable benchmark hinders our understanding of LLMs’ capabilities in visualization generation. In this paper, we address this gap by proposing a new NL2VIS benchmark called VisEval. Firstly, we introduce a high-quality and large-scale dataset. This dataset includes 2,524 representative queries covering 146 databases, paired with accurately labeled ground truths. Secondly, we advocate for a comprehensive automated evaluation methodology covering multiple dimensions, including validity, legality, and readability. By systematically scanning for potential issues with a number of heterogeneous checkers, VisEval provides reliable and trustworthy evaluation outcomes. We run VisEval on a series of state-of-the-art LLMs. Our evaluation reveals prevalent challenges and delivers essential insights for future advancements.