Telling Stories from Computational Notebooks: AI-Assisted Presentation Slides Creation for Presenting Data Science Work

Telling Stories from Computational Notebooks: AI-Assisted Presentation Slides Creation for Presenting Data Science Work

An opinionated summary

This paper introduces NB2Slides. NB2Slides is a JupyterLab extension for generating and editing presentation slides from Jupyter notebooks. NB2Slides is designed for Data Science and data scientists, while the presentation slides are for various audiences. NB2Slides generates the first draft, and data scientists are expected to iterate over it.

  • Authors: Chengbo Zheng, Dakuo Wang, April Yi Wang, Xiaojuan Ma
  • Year: 2022
  • Paper
  • HTML
  • Repo

NB2Slides interface

Takeaways

  • NB2Slides is a semi-automatic tool. There are automatically generated slides using Deep Learning and template-based slides for the data scientist to fill in. The data scientist can edit any slide from the UI.
  • It's not about full automation; it's about getting a head start and working collaboratively with AI. In other words, it is not possible to capture the data scientists' understanding of the problem and the business context. Also, current technology can only summarize what data scientists do but not why.
  • The slides follow an outline also defined in this paper.
  • Before generating the slides, two parameters can be adjusted: audience background and level of detail. In practice, the audience background parameter moves some content to appendices. It is unclear what the impact of the level of detail parameter is.
  • They conducted a formative study to understand how data scientists create presentation slides today. It consisted of 2-3 parts: literature review, creation of a presentation outline, and user interviews (think-aloud protocol)/participatory design sessions.
  • Data scientists or DS workers.
  • Notebook cells and slides are linked.
  • The authors have included an "Input Requirements" section listing best practices for using NB2Slides. NB2Slides requires high-quality notebooks as input.
  • NB2Slides includes a notebook minimap similar to the VS Code one. It is possible to show or hide the charts.
  • Some AutoML products, such as H2O.ai and DataRobot, have an automatic report generation feature. It is focused on documentation and not on the communication/presentation of a project.