Exploring Data Preparation Modules by Examples

When crafting data-analysis pipelines, a significant challenge lies in pinpointing the modules suited for data preparation, constituting 80% of such workflows. These versatile modules handle tasks like record retrieval, format transformation, and data combination. To aid scientists in module discovery, semantic annotations are valuable, yet they often fall short in fully describing module functionality. To address this, our paper proposes a solution combining semantic annotations with data examples to facilitate the discovery of data preparation modules. Our iterative discovery strategy enables scientists to explore existing modules via feedback on data examples. Evaluation exercises demonstrate the efficacy of our method, showcasing users’ ability to discover target modules with minimal feedback.