ComputableViz: Mathematical Operators as a Formalism for Visualization Processing and Analysis

Title of the Paper

ComputableViz: Mathematical Operators as a Formalism for Visualisation Processing and Analysis

Paper Information

• Research Domain: Data Visualization, Algorithms, Visual Processing
• Keywords: Data Visualization, Mathematical Operations, Visualization Analysis, Vega-Lite, Data Processing, Chart Manipulation, Multi-View, Data Pattern Study

Research Background and Problem Statement

• Identified Issues or Challenges:

  • With the rapid proliferation of data visualizations online, there is a growing need for processing and analyzing generated and digitized visualizations.
  • Current approaches predominantly focus on single visualizations, lacking a formal framework for operations across multiple visualizations, which complicates tasks like sorting and clustering.
  • The research field lacks a unified framework to organize concepts related to multiple data visualization operations.

• Significance:

  • Data visualization is rapidly becoming a distinct data format, and processing and analyzing these visualizations is crucial for understanding information and developing new methodologies. For instance, summarizing and organizing visualization information is critical in public health.
  • Visualization analysis serves as a gateway for users to access data, facilitating information comprehension and dissemination.

• Research Motivation and Related Work:

  • Inspired by mature methods in image processing and analysis, the authors aim to extend mathematical operations to the domain of data visualization.
  • Existing mathematical formalization methods for single visualization operations (e.g., Draco, Vega-Lite) lack mechanisms for operations across multiple visualizations.

Proposed Solution

• Proposed Method:

  • Introduced a formal framework for data visualization processing and analysis based on mathematical operations such as union, intersection, and difference.
  • Conducted systematic analysis of existing work to design a design space for visualization operations, implemented as a Python library named ComputableViz.
  • Leveraged Vega-Lite specifications and relational database principles to transform visualization specifications, enabling mathematical operations.

• Innovations:

  • First to incorporate mathematical operations into multi-visualization processing and analysis, abstracting complex visualization tasks (e.g., sorting, clustering) into simpler operations.
  • Modular implementation using relational database design to support diverse visualization tasks and promote code reuse.
  • Unified multiple research domains by integrating dispersed tasks such as style transfer, comparison, and clustering into a common framework.

• Implementation Steps and Key Techniques:

  1. Design Dimensions: Identify actionable targets (e.g., data, styles) and operation types (e.g., union, difference).
  2. Implementation Process:
     • Convert Vega-Lite specifications into relational database formats, including data tables, style tables, and mapping relationship tables.
     • Use relational algebra for operations (FULL OUTER JOIN for union, INNER JOIN for intersection, ANTI JOIN for difference).
     • Parameterize tasks for customization, such as style transfer and automatic encoding mechanisms.
  3. Validation: Validate the framework's applicability through various case studies and experiments.

Research Outcomes

• Specific Results:

  • Proposed a general framework capable of handling diverse tasks, with extensibility for new use cases.
  • Developed the ComputableViz library, providing automated and configurable solutions for multiple real-world applications.
  • Demonstrated use cases, including creating accessible visualizations, establishing visualization version control, analyzing gene relationship networks, and exploring chart clustering.

• Advantages:

  • A logically clear and easily extensible formal framework that enables unified modeling of different tasks.
  • Higher efficiency and automation compared to existing manual or complex models.
  • Modular methods defined by mathematical operations promote code reuse.

• Experimental or Evaluation Results:

  • Successfully performed style transfer on 374 Vega-Lite examples, with high adaptability in the test set.
  • In clustering experiments, the new embedding algorithm demonstrated higher separation in visual classification compared to the baseline GVAE method.

• Limitations and Future Directions:

  • Limitations:
    • The current version does not support operations on complex combinations and nested visualization specifications.
    • Limited automatic parsing capabilities for chart images, unable to handle noisy data or unstructured visual elements.
  • Future Directions:
    • Enhance methods for converting images to specifications to better support foundational visualizations widely present on the web.
    • Expand support for complex data semantics and cross-category analysis to improve tasks like visualization clustering and style transfer.
    • Develop an end-to-end visualization generation system based on the new formalism for advanced chart design (e.g., multi-view coordination and nested visualizations).

Through the authors' preliminary research, the field of data visualization processing and analysis is emerging as an independent and significant research direction. The authors provide a robust attempt at constructing a unified framework, paving the way for future studies.