Evaluating ChatGPT to support data visualization design

Abstract

Large Language Models (LLMs) like ChatGPT offer new avenues for diverse tasks, including complex activities such as data visualization design. While existing studies explore LLMs for specific design activities, particularly product generation and ideation, a comprehensive investigation into their support for the entire data visualization design process, especially for non-experts, remains largely unexplored. This work addresses this gap by investigating LLMs' capability to assist beginners in applying design methods and tools throughout the data visualization design process. Our research was guided by the question: “How can LLMs support the visualization design process?” A preliminary study explored prompt strategies for generating data visualization recommendations and established criteria for evaluating model response quality. Methodologically, we analysed design techniques, created prompts for non-specialist designers, and evaluated the process of building design guides with experts, defining a usage context and evaluation criteria for assessing user interaction and perception. Our findings indicate that ChatGPT can support both abstract and tangible design activities at varying levels. However, it is crucial that users possess sufficient domain knowledge to critically evaluate and refine the model’s outputs, ensuring accuracy and appropriateness in data visualization design.

Ähnliche Arbeiten