You can't always sketch what you want: Understanding Sensemaking in Visual Query Systems

Visual query systems (VQSs) allow users to interactively search for line charts with desired visual patterns typically specified using intuitive sketch-based interfaces. Despite their potential in accelerating data exploration, more than a decade of past work on VQSs has not been translated to adoption in practice. Through a year-long collaboration with experts from three diverse domains, we examine the role of VQSs in real data exploration workflows, develop features to support these workflows by enhancing an existing VQS, and evaluate how these features are used in practice. Via these observations, we formalize a taxonomy of key functionalities for VQSs, organized by three sensemaking processes. Perhaps somewhat surprisingly, we find that ad-hoc sketch-based querying is not commonly used during data exploration, since analysts are not able to precisely articulate the patterns they are interested in. We find that there is a spectrum of VQS-centric data exploration workflows, depending on the application, and that many of these workflows are not effectively supported in present-day VQSs. Our insights can pave the way for next-generation VQSs to be adopted in a variety of real-world applications.