Interactive visualisation of similarity structures

Abstract

In many operational contexts, data visualisation techniques are used to convey accurate and easily comprehended information to human analysts. One popular approach is to represent domains of interest in terms of the objects they contain and measures of the similarity between them. Spatial data visualisations can then be generated which place objects so that those which are more similar lie nearer each other. Unfortunately, this approach has limitations when dealing with domains which require the display of very high-dimensional configurations or incorporate fundamentally non-metric structures. This paper develops and demonstrates an interactive approach to spatial data visualisation which addresses both of these weaknesses. By allowing analysts to select objects within the domain and view spatial configurations adjusted to represent the similarity relationships from the perspective of those objects, it is shown to be possible to build up an accurate and useful understanding of high-dimensional and non-metric domains.