Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Exploring the Capability of LLMs in Performing Low-Level Visual Analytic Tasks on SVG Data Visualizations

Zhongzheng Xu, Emily Wall

TL;DR

This study investigates whether large language models (LLMs) can execute Amar, Eagan, and Stasko's 10 low-level visual analytic tasks directly on scalable vector graphics (SVG) visualizations using zero-shot prompts. By evaluating three chart types (scatterplot, line chart, bar chart) across two dataset sizes and two labeling schemes with 400 stimuli, the authors measure task success with exact-match accuracy. Findings show that LLMs reliably perform pattern-based tasks like clustering, but struggle with arithmetic and complex operations such as Compute Derived Value and Correlate, with performance influenced by data point count and labeling. The work highlights both the potential and current limitations of applying LLMs to direct SVG manipulation for visual analytics, and suggests integrating scripting or execution tools and expanding tested scenarios in future research to realize practical applications.

Abstract

Data visualizations help extract insights from datasets, but reaching these insights requires decomposing high level goals into low-level analytic tasks that can be complex due to varying degrees of data literacy and visualization experience. Recent advancements in large language models (LLMs) have shown promise for lowering barriers for users to achieve tasks such as writing code and may likewise facilitate visualization insight. Scalable Vector Graphics (SVG), a text-based image format common in data visualizations, matches well with the text sequence processing of transformer-based LLMs. In this paper, we explore the capability of LLMs to perform 10 low-level visual analytic tasks defined by Amar, Eagan, and Stasko directly on SVG-based visualizations. Using zero-shot prompts, we instruct the models to provide responses or modify the SVG code based on given visualizations. Our findings demonstrate that LLMs can effectively modify existing SVG visualizations for some tasks like Cluster but perform poorly on tasks requiring mathematical operations like Compute Derived Value. We also discovered that LLM performance can vary based on factors such as the number of data points, the presence of value labels, and the chart type. Our findings contribute to gauging the general capabilities of LLMs and highlight the need for further exploration and development to fully harness their potential in supporting visual analytic tasks.

Exploring the Capability of LLMs in Performing Low-Level Visual Analytic Tasks on SVG Data Visualizations

TL;DR

This study investigates whether large language models (LLMs) can execute Amar, Eagan, and Stasko's 10 low-level visual analytic tasks directly on scalable vector graphics (SVG) visualizations using zero-shot prompts. By evaluating three chart types (scatterplot, line chart, bar chart) across two dataset sizes and two labeling schemes with 400 stimuli, the authors measure task success with exact-match accuracy. Findings show that LLMs reliably perform pattern-based tasks like clustering, but struggle with arithmetic and complex operations such as Compute Derived Value and Correlate, with performance influenced by data point count and labeling. The work highlights both the potential and current limitations of applying LLMs to direct SVG manipulation for visual analytics, and suggests integrating scripting or execution tools and expanding tested scenarios in future research to realize practical applications.

Abstract

Data visualizations help extract insights from datasets, but reaching these insights requires decomposing high level goals into low-level analytic tasks that can be complex due to varying degrees of data literacy and visualization experience. Recent advancements in large language models (LLMs) have shown promise for lowering barriers for users to achieve tasks such as writing code and may likewise facilitate visualization insight. Scalable Vector Graphics (SVG), a text-based image format common in data visualizations, matches well with the text sequence processing of transformer-based LLMs. In this paper, we explore the capability of LLMs to perform 10 low-level visual analytic tasks defined by Amar, Eagan, and Stasko directly on SVG-based visualizations. Using zero-shot prompts, we instruct the models to provide responses or modify the SVG code based on given visualizations. Our findings demonstrate that LLMs can effectively modify existing SVG visualizations for some tasks like Cluster but perform poorly on tasks requiring mathematical operations like Compute Derived Value. We also discovered that LLM performance can vary based on factors such as the number of data points, the presence of value labels, and the chart type. Our findings contribute to gauging the general capabilities of LLMs and highlight the need for further exploration and development to fully harness their potential in supporting visual analytic tasks.
Paper Structure (5 sections, 3 figures, 1 table)

This paper contains 5 sections, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Result
  5. Discussion and Conclusion

Figures (3)

  • Figure 1: Illustration of the study design (top) and examples of the Cluster and Determine Range tasks (bottom).
  • Figure 2: An example of the prompt for the task Find Anomalies.
  • Figure 3: Comparison of retrieved data points (teal) extracted by the LLM directly from SVG position definitions and the original data points (orange).