Aperture DemonstrationsThe following apps were developed as part of a series of exercises in agile development of visual analytics for big data. The exercises, each two to four weeks, were designed to test rapid development of tailored solutions for complex analytic problems using Aperture, and to use the results to inform further development of the library.
Money Flow: CharityNetCharityNet looks at the problem of large node-link data sets (graphs). In big data typical of many real world problems, a brute force approach of showing every node in the data set does not scale (Schneiderman, 2006, et al). Issues include inability to support user questions and task functions, limited visual expression of node and link semantics, and undue effort required to explore and understand.
Our research objectives in this phase were to investigate and develop aggregated graph visualization techniques that:
- Effectively summarize communities of nodes and their relationships.
- Do not sacrifice key member node characteristics in the aggregation.
- Are immediately readable, not requiring interaction for basic sense-making.
- Graph Clustering, using Flow, Community, and Multi-Attribute techniques.
- Aggregation Markers, using "Community Rings" to summarize member nodes.
- Adaptive Labeling, by applying an optimized "Trellis Strategy" (Mote, 2007).
The Analytic QuestionCharityNet consists of a data set of 5,730 charities, 1.8M donors, and 3.3M donations over 2 years. The decision-oriented, analytic question posed for design of this exercise's solution was: Reveal a charity's donor characteristics, and find strategies for increasing support.
Behavioral Trends and Projections, with TransparencyThis monitoring application displays the past and projected behavior of key indicators, starting with a geospatial summary across all tracked countries. Clicking on a country of interest shows its indicators in detailed time series form. Further drill-down into any one indicator reveals influence of contributing factors.
Transparency of the underlying model is provided by a visualization technique of linked visible behaviors, coupled with narrative explanation. Provenance of source data (not available in the model shown) is provided by reference annotations.
This app was developed over a period of several months with a very early version of Aperture. The hybrid computational model and source data consisted of up to twelve years of weekly and annual values for hundreds of variables, feeding key indicators for each of five countries. Reference annotations provided a visualization of volume and provenance of data, with links to source data that included key-word markup of articles mined for sentiment analysis.
The model and data set shown here is a more limited example, built on a model of the steel industry. Not all features are supported by this model.