Progressive Visualization for Big Data Exploratory Analysis

Abstract

We live in a world where our ability to generate data far exceeds our ability to analyze it. Visual analytics systems have become a key decision making tool in data analysis, helping users gain an understanding of large datasets through sophisticated data management and performing computationally intensive tasks. Yet often these tasks can take minutes, hours or even days to complete, preventing users from interacting with their data in an efficient manner. Progressive visualization is fast becoming a technique in the visualization community to help users interact with large amounts of data. With progressive visualization, users can examine intermediate results of complex or long running computations, without waiting for the computation to complete.

However, with progressive visualization still a nascent analysis technique, a paradigm shift is needed in the way of existing computational algorithms, databases and visual interfaces. There is also the need to examine how progressive visualization affects a human user and their decision making. The work in this dissertation evaluates multiple aspects of progressive visualization as a data analysis technique. First, by presenting a data sampling method for quickly improving estimates of queries commonly used in visualizations. This sampling method also enables users to steer the progression such that parts of the visualization they are more interested in can reach higher fidelity estimates faster. Next, by performing user studies to evaluate different interface designs for presenting the uncertainty associated with intermediate computational results, as well as effective steering controls. These studies also include how different steering control designs can alleviate the cognitive load of steering while the progression occurs. Following this, a progressive visualization system is developed combining the new data sampling method with the lessons learned from the interface design studies and then evaluated with expert users. To evaluate how progressive visualization affects users, we examine four cognitive biases that can be present during the analysis: Uncertainty Bias, Illusion Bias, Control Bias and Anchoring Bias, as well as the advantages and limitations of progressive visualization. Finally, we discuss how fast response times are not necessarily the ultimate goal in a visual analytics system if there is an opportunity to improve the user's decision making.

Join the meeting in Sococo VH 209. Login: tuftscs.sococo.com

Zoom Location: https://tufts.zoom.us/my/marianneprocopio

Password: see colloquium email

Dial-in option not available.