Visualizationary: Automating Design Feedback for Visualization Designers Using LLMs

Area of Science:

• Computer Science
• Human-Computer Interaction
• Data Visualization

Background:

• Interactive visualization editors lack guidance on effective visual communication.
• Large language models (LLMs) offer potential for providing design feedback.

Purpose of the Study:

• To explore using LLMs for customized feedback in visualization design.
• To evaluate the impact of LLM-generated guidance on visualization refinement.

Main Methods:

• Developed Visualizationary, integrating ChatGPT with design guidelines and perceptual filters.
• Conducted a longitudinal user study with 13 visualization designers of varying expertise.
• Collected data on the design process and visualization outcomes over several days.

Main Results:

• LLM-generated natural language feedback aided designers in refining visualizations.
• Guidance proved beneficial across novice, intermediate, and expert skill levels.
• Participants found the AI-driven feedback actionable for improving visual communication.

Conclusions:

• LLMs can serve as effective tools for providing personalized feedback in visualization design.
• Integrating AI into design tools can democratize effective visual communication principles.
• Future work can expand LLM capabilities for more sophisticated design assistance.