Date of Original Version
Abstract or Table of Contents
Exploratory data analysis is a human-centered process whose goal is to develop intuition and insight into a dataset. Usually initial findings will suggest new questions, leading to extensive iteration. Statistical graphics play an important role, but a flexible means for populating them with relevant data is also needed. Database query languages are one means, but formulating queries is a distraction from the primary task of understanding the data. This paper uses the concept of isets, which embody both a description and a set of entities that satisfy it, to bridge the gap between the expressive power of queries and the concreteness of data. During exploration, isets can be incrementally modified, related to other isets, or used to define new isets. This last operation uses the metaphor of navigation from an existing iset as the means of arriving at the new related iset, which situates the analyst as actively moving through the world of the data. The navigation path among isets resembles a graphical query language. However, visualizations give continuous feedback as the graph is incrementally constructed, rather than only after an “execute” command. Navigation and nine other operations are meant to effectively support iterative exploration.