InkSight: Leveraging Sketch Interaction for Documenting Chart Findings in Computational Notebooks.

Area of Science:

• Computer Science
• Data Visualization
• Human-Computer Interaction

Background:

• Computational notebooks are popular for data analysis, but documenting chart findings is challenging.
• Existing automatic methods for documentation often lack user specificity.
• Effective documentation is crucial for recalling and sharing data analysis insights.

Purpose of the Study:

• To develop InkSight, a mixed-initiative plugin for computational notebooks.
• To enable users to express intent for data subsets via intuitive sketching.
• To automate the generation of chart finding documentation tailored to user interests.

Main Methods:

• InkSight uses open-path and closed-path sketches to identify user intent and data subsets.
• It filters data fact types based on sketches and selected data.
• Large language models (GPT-3.5) convert inferred data facts into natural language documentation.

Main Results:

• InkSight successfully generates relevant documentation based on user sketches.
• Users can intuitively express their intent through sketching on visualizations.
• The generated documentation can be conveniently fine-tuned by users.

Conclusions:

• InkSight enhances the process of documenting chart findings in computational notebooks.
• The plugin effectively bridges the gap between user intent and automated documentation.
• User studies confirm InkSight's usability and effectiveness in facilitating data analysis documentation.