Establishing a Cooperation Between RadViz and SOM to Improve the Analyst Visual Experience

Abstract

Radial Visualization (RadViz) is an information visualization technique for visual data exploration and data analysis. RadViz transforms multidimensional data into two dimensional features. A circle is drawn and, then, respectively, dimensions and instances become axes and points. It aims to enable human identification of similarities among instances stored in datasets. RadViz has a dimension arrangement problem since different radial axes disposal produce different point arrangements misleading, the expert user, about the most appropriate for interpretation. In this work, the results produced by an automated method (Self Organizing Maps (SOM)) and a visual metaphor (RadViz) are committed into a strict cooperation through a middle layer. The middle layer main objective is to identify an appropriate dimension arrangement. We propose that the dimension arrangement should be computed at the multidimensional space by analyzing reference vectors produced by SOM. The reference vectors are analyzed so that similar dimensions become neighbors. Later the achieved dimension arrangement is communicated to RadViz for rendering proposes and used to render the instances. Expert users can analyze all neurons, a combination of neurons or isolated neurons. The dimension arrangement represents a relational and universal order established for dimensions. In summary, the system inspects and proposes axes arrangement on-the-fly as users adds and removes neurons and dimensions.