Abstract:
|
At Facebook more than 1.4 billion monthly active users generate amounts of data that test the limits of data infrastructures, statistical tools, and methods for data visualization. Insights can often be gained by using sampled data. However, when visualizing large-scale spatio-temporal processes, especially in cases in which high-level patterns are different from low-level patterns across space and time, letting viewers interactively explore different levels of aggregation necessitates working with the full dataset. This in turn often requires either pre-rendering of tiles or using techniques such as Nanocubes. I will present some approaches for characterizing the necessary level of detail required to meaningfully explore such data as well as techniques to allow for interactive exploration.
|