“…A body of related work proposed techniques for sub-tasks in interactive visualization recommendation systems, such as improving expressiveness or perceptual effectiveness [24], matching user intents [10], displaying trends and outliers [20], etc. These sub-tasks can generally be divided into two categories [18,42]: whether the solution focuses on recommending data (what data to visualize), such as Discovery-driven Data Cubes [33], Scagnostics [39], Auto-Vis [40], AutoPartition [2], Foresight [6], SpotLight [11], ScagExplorer [5], VisPilot [19], VizDeck [28] and MuVE [9] or recommending encoding (how to design and visually encode the data), such as APT [24], ShowMe [25], BDVR [10], SeeDB [38], Draco-learn [26], and LQ 2 [44]. While some of those are ML-based [14,19,26,28], none recommends entire visualizations, and thus does not solve the visualization recommendation problem that lies at the heart of our work.…”