TreeView: A Structured User Interface for Understanding, Verifying, and Extending Unfamiliar Computational Notebooks

Computational notebooks are widely used in data science for their interactive, flexible environment. However, their linear format can hinder comprehension, verification, and extension, especially when users engage with unfamiliar or poorly documented notebooks. We present TreeView, a JupyterLab extension that transforms the notebook into a semantically structured, execution-aware tree-like structure. By combining automatic exploratory data analysis (EDA) stage categorization, cell grouping and summarization, inline execution/output inspection, and natural language querying, TreeView provides a unified interface for high-level reasoning and in situ exploration. Through a controlled within-subjects study (N = 14), participants using TreeView completed tasks faster than with the traditional linear notebook baseline and achieved higher accuracy when identifying appropriate extension points. Participants also reported lower cognitive load and greater satisfaction, citing improved orientation, verification, and workflow continuity. Based on these findings, we discuss design insights for intelligent interfaces that support short-term sense-making and navigation in unfamiliar computational notebooks.