[Visinfo] Week 8 Skupin

[Corina S. Schweller]

"A Cartographic Approach" Skupin

Skupin discusses the improvement of map-like visualizations of 
information that is not geographic. He creates visualizations that 
resemble maps and are meant to be read as we read maps. He suggests 
using hierarchical clustering to group features and labeling based on 
scale to make the maps readable. Geometric proximity will show 
similarity in his Base Map. Creating a keyword index that is automated 
is the first step. The vector-space model shows the connection between 
keywords and the document. Self-organizing maps (Kohonen Maps)  have 
the ability to show large amounts of information and is scalable and 
trainable so that relationships between inputs are conserved. For 
scale-dependant information the best choice is hierarchical clustering, 
such as clustering trees. What is important to keep in mind while 
choosing a clustering method is that the amount of graphic detail shown 
can be controlled. Another important think id the labeling of 
individual documents and document clusters. Naming a document comes 
from the most prevalent key word found in the document. Traditional 
cartographic design can use the data sets of the computation of labels 
and merging of base-map polygons. These concepts can be used for 
conference mapping and Skupin believes that they can be placed in a 
fully automated system that visualy show conference abstracts.