Abstract
Relating data across multiple visualizations has been a prevalent task for visual analytics in various fields (e.g., bioinformatics, business, and cybersecurity). For example, a marketing analyst needs to relate periods from a line chart of car sales revenue with nodes in a scatterplot of cars by price and horsepower, and cells from a matrix that summarizes transaction amounts by types of deals. Currently displaying data relationships across multiple visualizations heavily relies on coordination-based techniques (e.g., brushing and linking), which requires significant user effort (e.g., many trial-and-error attempts) to explore and see cross-view data relationships. Thus, how to design techniques for supporting cross-visualization relationship explorations remains a challenging problem. In order to address this, we propose a novel technique, highlighting a context-separation design concept, which displays computed cross-view data relationships in a stand-alone visualization. We have developed a prototype software, as a proof of concept, and performed a user experiment to study the impact of our proposed technique on exploring data relationships across multiple visualizations. Our study shows that the data relationship view performs better than the stand-alone view in terms of accuracy of user findings which involves relationship comparison and has an almost similar number of user interactions considering almost double the number of nodes in data relationship view.