An Analytic Approach for Discovery.
Eric Dull, Steven P. Reinhardt · 2014
Abstract—With the widespread awareness of Big Data, mission leaders now expect that the data available to their organizations will be immediately relevant to their missions. However, the continuing onslaught of the ”data tsunami”, with data becoming more diverse, changing nature more quickly, and growing in volume and speed, confounds all but the simplest analysis and the most capable organizations. The core challenge faced by an analyst is to discover the most important knowledge in the data. She must overcome potential errors and inaccuracies in the data and explore it, even though she understands it incompletely, guided by her knowledge of the data’s domain. We have solved customer problems by quickly analyzing numerous dimensions of data to check its sanity and to compare it to expected values. Guided by what we find initially, we quickly move on to further (unanticipated) dimensions of the data; discovery depends on this ability. This approach vitally brings the analyst into direct interaction with the data. We implement this approach by exploiting the ability of graphs (vertices and edges) to represent highly heterogeneous data. We use RDF as the data representation and SPARQL for queries. We use non-parametric tests, which are readily targeted to any type of data. This graph-analytic approach, using proven techniques with widely diverse data, represents a guidepost for delivering greatly improved analysis on much more complex data. I.