Large Multiples : exploring the large-scale scattergun approach to visualization and analysis

Abstract

We create 2.5 quintillion bytes of data every day. A whole 90% of the world’s data was created in the last two years.1 One contribution to this massive bulk of data is Twitter: Twitter users create 500 million tweets a day,2 which fact has greatly impacted social science [24] and journalism [39]. Network analysis is important in social science [6], but with so much data there is a real danger of information overload, and there is a general need for tools that help users navigate and make sense of this. Data exploration is one way of analyzing a data set. Exploration-based analysis is to let the data suggest hypotheses, as opposed to starting out with a hypothesis to either confirm or refute. Visualization is an important exploration tool. Given the ready availability of large-scale displays [1], we believe that an ideal visual exploration system would leverage these, and leverage the fact that there are many different ways to visualize something. We propose to use wall- sized displays to provide many different views of the same data set and as such let the user explore the data by exploring visualizations. Our thesis is that a display wall architecture [1, 42] is an ideal platform for such a scheme, providing both the resolution and the compute power required. Proper utilization of this would allow for useful sensemaking and storytelling. To evaluate our thesis we have built a system for gathering and analyzing Twitter data, and exploring it through multiple visualizations. Our evaluation of the prototype has provided us with insights that will allow us to create a practicable system, and demonstrations of the prototype has uncovered interesting stories in our case study data set. We find that it is strictly necessary to use clever pre-computation, or pipelining, or streaming to meet the strict latency requirements of providing visualization interactively fast. Our further experiments with the system have led to new discoveries in streaming graph processing.