Exploratory Data Analytics for Information Discovery in a Network Structure

摘要

This report presents an analytic strategy for visual exploration of multidimensional data. Node position in a network structure is determined by projecting from the high-dimensional data (HDD) space to a low-dimensional latent space. Clustering of node position vectors may result for making inferences. Dimensionality reduction by feature extraction of HDD for visualization is performed using a parametric Student's t-distribution for stochastic neighbor embedding (t-SNE). The resultant t-SNE network of nodes for a Euclidean space can now be examined using visual analytics technology- navigation/interaction within the visualization of the data. Scene content is described using the Extensible 3-D (X3D) graphics application programming interface. The immersive profile of an X3D scene allows for navigation within the data for possible information discovery. Such an approach may provide for a better understanding of data and facilitate analytical reasoning that would otherwise be difficult in an exclusively textual context.