Design and Evaluation of Spatialized 2D Computational Notebooks for Data Science

Abstract

Computational notebooks are a popular tool for data science due to their ability to interleave text documentation, code, and results, including visuals, into a computational narrative. However, they have certain limitations which may be caused or exacerbated by the linear, top-down, 1D organization of cells within them. This dissertation explores and evaluates the potential of nonlinear computational notebooks to address issues with current linear computational notebooks, such as inefficient use of larger display spaces, and improve on the state of computational notebooks generally for data science. In this dissertation, we performed several studies aimed at exploring and evaluating the potential of spatialized 2D computational notebooks. We found in our first study several patterns of nonlinearity in data science work with computational notebooks and evaluated how problematic each pattern is. The second study found that users would utilize 2D space to organize computational notebook cells and discovered several patterns of organization. The third study showcased how comparative analysis can be more efficient and easier in well-designed spatialized 2D computational notebooks. The final study found that freeform spatialized 2D computational notebooks were seen as more usable than linear computational notebooks. Overall, our studies helped begin the work of expanding computational notebooks beyond their current linear mold.