Evaluating LLMs for Visualization Generation and Understanding

TL;DR

This paper evaluates the ability of large language models to generate visualization code and understand visualizations, highlighting their strengths with simple tasks and limitations with complex visualizations and detailed questions.

Contribution

It provides an empirical assessment of LLMs' capabilities in visualization generation and understanding, revealing their potential and current limitations.

Findings

LLMs can generate code for simple visualizations like bar and pie charts.

LLMs can answer basic questions about visualizations.

Struggles with complex visualizations and detailed relational questions.

Abstract

Information Visualization has been utilized to gain insights from complex data. In recent times, Large Language models (LLMs) have performed very well in many tasks. In this paper, we showcase the capabilities of different popular LLMs to generate code for visualization based on simple prompts. We also analyze the power of LLMs to understand some common visualizations by answering questions. Our study shows that LLMs could generate code for some simpler visualizations such as bar and pie charts. Moreover, they could answer simple questions about visualizations. However, LLMs also have several limitations. For example, some of them had difficulty generating complex visualizations, such as violin plot. LLMs also made errors in answering some questions about visualizations, for example, identifying relationships between close boundaries and determining lengths of shapes. We believe that our insights can be used to improve both LLMs and Information Visualization systems.