Alternative Visualization Methods to Explore Larger Dendrograms

Gilles Bisson (1) , Renaud Blanch (2)

1 AMA - Analyse de données, Modélisation et Apprentissage automatique [Grenoble]
2 IIHM - Ingénierie de l’Interaction Homme-Machine

In this paper, we present a new hybrid visualization method based on the integration of two recent approaches: 'Treematrix' and 'Stacked Trees'. Our objectives are twofold. On the one hand, it allows to display large trees containing several thousand of instances in an uncluttered and scalable way. Moreover, different information can be simultaneously displayed about each instances without the need of using filter or zoom. On the other hand, by using a tree-matrix based interactive visualization of dendrograms, we enable users to do tasks involving both clusters and individual objects that are impracticable with the classical representation. For example: to explain why a specific object belongs to a specific cluster or to elicit and understand uncommon patterns such as instances related to several clusters.

Domaines: Apprentissage [cs.LG]; Interface homme-machine [cs.HC]; Machine Learning [stat.ML]

Gilles Bisson, Renaud Blanch. Alternative Visualization Methods to Explore Larger Dendrograms. In proceedings of the European Conference on Data Analysis, 2013, France. pp.1-1. ⟨hal-00952838⟩