Charagraph: Interactive Generation of Charts for Realtime Annotation of Data-Rich Paragraphs

Document Title

Charagraph: Interactive Generation of Charts for Realtime Annotation of Data-Rich Paragraphs

Document Information

• Subject Area: Human-Computer Interaction, Data Visualization, and Interactive Document Tools
• Keywords: Visualization System, Reading Interface, Data-Rich Documents, Dynamic Interaction, Realtime Annotation, Data Analysis, Document Augmentation

Research Background and Problem

• Identified Issues or Challenges:

  • Scientific literature and reports often contain extensive numerical data embedded within sentences to support arguments, making it difficult to compare, interpret, and integrate such data.
  • Presenting numerical data linearly in text limits readers' ability to uncover data patterns and identify trends.
  • Even when documents include visualizations, readers may overlook them due to design issues or their independent placement.

• Why It Matters:

  • The readability of numerical information and its correlation with visualizations are critical for scientific interpretation.
  • Enhancing data comprehension through interactive tools without modifying the original document can improve the reading experience and support decision-making.

• Research Motivation and Related Work:

  • Current solutions focus on linking existing graphics with text or creating interactive documents directly by authors, with limited research on extracting data from sentences to generate visualizations.
  • The need for automated visualization tools is particularly pressing when data is embedded in textual sentences rather than tables.

Solution

• Proposed Method or Solution:

  • Introduced the concept of "Charagraph," which enables readers to dynamically generate interactive charts while reading, facilitating the visualization, comparison, and manipulation of numerical data within text.

• Innovative Features:

  1. Extract numerical data from semantically rich text to generate visualizations.
  2. Provide interactive features that tightly integrate text and charts, enabling bidirectional synchronization.
  3. Allow readers to dynamically merge chart information, such as combining series data or adding error bars.

• Implementation Steps and Techniques:

  1. Text Selection and Data Extraction:
     • Supports traditional text selection or rectangular selection tools to extract numerical data from PDFs.
     • Utilizes natural language features (e.g., units, statistical terms, context) to generate data grouping suggestions.
  2. Chart Customization:
     • Supports various common chart types, including bar charts, line charts, and pie charts, with dynamic adjustments for labels and scales.
  3. Information Interaction and Integration:
     • Enables users to sort, filter, highlight, and compare chart values.
     • Provides functionality to integrate multiple data parts into a single chart, such as adding series or error bars.
  4. System Implementation:
     • Built using TypeScript and React, with chart rendering powered by Apache ECharts, and integrated with a PDF.js-based document reader.

Research Outcomes

• Specific Outcomes:

  • Developed and validated an interactive document tool that allows readers to dynamically generate Charagraph charts from PDF documents.
  • User experiments demonstrated that the tool enhances the readability of data-intensive text and improves the accuracy of readers' responses to related questions.

• Advantages Compared to Existing Solutions:

  • Does not rely on predefined tables or author-provided chart templates, enabling readers to generate highly customized visualizations based on their needs.
  • Offers various interactive features such as dynamic highlighting and realtime comparison, enhancing data exploration capabilities.

• Experimental or Evaluation Results:

  • A user study (12 participants) showed that the accuracy of tasks completed using Charagraph was higher than relying solely on text (98.61% vs. 92.59%).
  • While task completion time did not significantly improve, the tool notably reduced readers' cognitive load, operational barriers, and frustration.
  • NASA-TLX questionnaire evaluations indicated that Charagraph significantly reduced perceived mental workload and effort (p<.003).

• Limitations and Future Directions:

  • Limitations:
    • Currently lacks support for non-numerical data in text (e.g., "half of the participants").
    • Requires more advanced NLP technologies to further reduce the need for manual data organization by users.
  • Future Directions:
    • Expand visualization types, such as scatter plots and histograms.
    • Provide more efficient natural language processing (NLP) to enhance data grouping accuracy and chart generation intelligence.
    • Explore the potential for implementing Charagraph in physical documents, such as through augmented reality technologies.