VICO: Visualizing Connected Object Orderings

Abstract

In modern databases, complex objects like multimedia data, proteins or text objects can be modeled in a variety of representations and can be decomposed into multiple instances of simpler sub-objects. The similarity of such complex objects can be measured by a variety of distance functions. Thus, it quite often occurs that we have multiple views on the same set of data objects and do not have any intuition about how the different views agree or disagree about the similarity of objects. VICO is a tool that allows a user to interactively compare these different views on the same set of data objects. Our system is based on OPTICS, a density-based hierarchical clustering algorithm which is quite insensitive to the choice of parameters. OPTICS describes a clustering as a so-called cluster order on a data set which can be considered as an image of the data distribution. The idea of VICO is to compare the position of data objects or even complete clusters in a set of data spaces by highlighting them in various OPTICS plots. Therefore, VICO allows even non-expert users to increase the intuitive understanding of feature spaces, distance functions and object decompositions.