From Interaction to Intent: A Crowdsourced Provenance Dataset for Multidimensional Projection Exploration

Crowdsourced hover-based interaction provenance from 263 non-expert participants performing seven multidimensional-projection (MDP) exploration tasks across four datasets and three projection methods (PCA, t-SNE, UMAP). Each of the 1,437 records is a task-level hover sequence — the ordered points a participant hovered on while solving one analytical task — with the underlying source data and 2D projection coordinates needed to reconstruct each visual layout. Data are de-identified (sequential participant IDs, per-sequence relative timestamps). This is the dataset companion to the paper "From Interaction to Intent: Inferring User Objectives from Provenance Logs" (arXiv:2607.04501). Reproduction code: https://github.com/5teffen/interaction-to-intent. See README.md and CODEBOOK.md for schema and collection methodology (ETH Zurich ethics ETHICS-413).