Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DeepVIS: Bridging Natural Language and Data Visualization Through Step-wise Reasoning

Zhihao Shuai, Boyan Li, Siyu Yan, Yuyu Luo, Weikai Yang

TL;DR

This work tackles transparency and refinement challenges in NL2VIS by integrating Chain-of-Thought reasoning into the NL2VIS pipeline. It introduces nvBench-CoT to capture explicit step-by-step reasoning and DeepVIS, an interactive interface that exposes and lets users edit the reasoning chain, enabling targeted refinements. Quantitative results show state-of-the-art All Acc and strong axis/data accuracy, while user studies confirm enhanced trust and usability through reasoning transparency. Overall, the approach advances human–AI collaboration in visualization by coupling automated generation with interpretable, controllable reasoning processes across data-to-visualization tasks.

Abstract

Although data visualization is powerful for revealing patterns and communicating insights, creating effective visualizations requires familiarity with authoring tools and often disrupts the analysis flow. While large language models show promise for automatically converting analysis intent into visualizations, existing methods function as black boxes without transparent reasoning processes, which prevents users from understanding design rationales and refining suboptimal outputs. To bridge this gap, we propose integrating Chain-of-Thought (CoT) reasoning into the Natural Language to Visualization (NL2VIS) pipeline. First, we design a comprehensive CoT reasoning process for NL2VIS and develop an automatic pipeline to equip existing datasets with structured reasoning steps. Second, we introduce nvBench-CoT, a specialized dataset capturing detailed step-by-step reasoning from ambiguous natural language descriptions to finalized visualizations, which enables state-of-the-art performance when used for model fine-tuning. Third, we develop DeepVIS, an interactive visual interface that tightly integrates with the CoT reasoning process, allowing users to inspect reasoning steps, identify errors, and make targeted adjustments to improve visualization outcomes. Quantitative benchmark evaluations, two use cases, and a user study collectively demonstrate that our CoT framework effectively enhances NL2VIS quality while providing insightful reasoning steps to users.

DeepVIS: Bridging Natural Language and Data Visualization Through Step-wise Reasoning

TL;DR

This work tackles transparency and refinement challenges in NL2VIS by integrating Chain-of-Thought reasoning into the NL2VIS pipeline. It introduces nvBench-CoT to capture explicit step-by-step reasoning and DeepVIS, an interactive interface that exposes and lets users edit the reasoning chain, enabling targeted refinements. Quantitative results show state-of-the-art All Acc and strong axis/data accuracy, while user studies confirm enhanced trust and usability through reasoning transparency. Overall, the approach advances human–AI collaboration in visualization by coupling automated generation with interpretable, controllable reasoning processes across data-to-visualization tasks.

Abstract

Although data visualization is powerful for revealing patterns and communicating insights, creating effective visualizations requires familiarity with authoring tools and often disrupts the analysis flow. While large language models show promise for automatically converting analysis intent into visualizations, existing methods function as black boxes without transparent reasoning processes, which prevents users from understanding design rationales and refining suboptimal outputs. To bridge this gap, we propose integrating Chain-of-Thought (CoT) reasoning into the Natural Language to Visualization (NL2VIS) pipeline. First, we design a comprehensive CoT reasoning process for NL2VIS and develop an automatic pipeline to equip existing datasets with structured reasoning steps. Second, we introduce nvBench-CoT, a specialized dataset capturing detailed step-by-step reasoning from ambiguous natural language descriptions to finalized visualizations, which enables state-of-the-art performance when used for model fine-tuning. Third, we develop DeepVIS, an interactive visual interface that tightly integrates with the CoT reasoning process, allowing users to inspect reasoning steps, identify errors, and make targeted adjustments to improve visualization outcomes. Quantitative benchmark evaluations, two use cases, and a user study collectively demonstrate that our CoT framework effectively enhances NL2VIS quality while providing insightful reasoning steps to users.

Paper Structure

This paper contains 28 sections, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Chain-of-Thought (CoT)
  4. NL2VIS
  5. Visualization for Human-LLM Collaboration
  6. Design of the CoT Process for NL2VIS
  7. Literature Review and Expert Interviews
  8. Initial Three-Stage Process
  9. Refined to Five-Stage CoT Process for NL2VIS
  10. Validation of the Five-Stage Process
  11. Construction of nvBench-CoT
  12. Database Description Module
  13. Reasoning Steps Generation Module
  14. nvBench-CoT
  15. DeepVIS
  16. ...and 13 more sections

Figures (5)

  • Figure 1: The pipeline of DeepVIS Framework
  • Figure 2: The interface of DeepVIS. The CoT view (A) provides a structured overview of the reasoning process and allows users to interactively refine it, while three supporting views (B-D) complement the CoT view during the analysis.
  • Figure 3: Comparative analysis of VQLs and generated charts between our method and baseline methods.
  • Figure 4: After using manual correction to specify the preference of the chart type, DeepVIS correctly updates the subsequent reasoning results and generates a proper aggregated visualization.
  • Figure 5: User study results: DeepVIS has consistently received relatively higher ratings across all six dimensions.