VizAgent: Towards an Intelligent and Versatile Data Visualization Framework Powered by Large Language Models

EasyChair Preprint 15668

Abstract

This study proposed VizAgent, a novel framework designed to address the challenges in data visualization. By leveraging the capability of Large Language Model (LLM), VizAgent automates describe data, guides users' intentions, automatically generates visualizations for specific tasks, and compares the quality of generated visualizations across different libraries. The results demonstrated that the matplotlib library outperformed other visualization libraries in terms of success rate, suggesting opportunities for further investigation and improvement, particularly in enhancing the performance of sea-born, plotly, and ggplot visualizations. The VizAgent framework presents a promising approach to intelligent data visualization, with several avenues for extending its capabilities. These include specialized handling for certain visualization libraries, empowering users to fine-tune parameters and styling, and incorporating advanced data analysis and feature engineering capabilities. VizAgent contributes to the ongoing efforts in data visualization by providing a valuable resource for researchers, practitioners, and individuals seeking data-driven decision-making.

Keyphrases: Generative AI, LLM, Python, Streamlit, data visualization

@booklet{EasyChair:15668,
  author    = {Hue Luong Thi Minh and Vinh Nguyen The and Truong Quach Xuan},
  title     = {VizAgent: Towards an Intelligent and Versatile Data Visualization Framework Powered by Large Language Models},
  howpublished = {EasyChair Preprint 15668},
  year      = {EasyChair, 2025}}