ANGULAR TREEMAPS - Information Visualisation: Techniques, Usability and Evaluation - page 85

generation for object recognition and change tracking. For example,
embedding distinctive shapes in traditional rectangular Treemaps can
highlight anomalies.
Motivated by this idea, we propose a new space-filling visualization
method named Angular Treemaps . These treemaps are capable of laying
out large relational structures and highlighting important substructures.
This new method uses the Divide and Conquer approach and Treemap
paradigm to represent and emphasize hierarchical information. It further
adopts the polygon-based enclosure partition to interactively subdivide the
display space into sub-polygons. The variation created by the angular
partitioning algorithm helps users to locate and identify differentiation.
This technique can be used manually or be applied by means of automatic
data mining. We now discuss related work in detail.
Related Work
Treemaps, first proposed by Johnson & Shneiderman in 1991, used the
space-filling partitioning concept to represent hierarchical structures
within the rectangular display space [9]. The general algorithm is to
recursively enclose nodes of the tree in rectangular areas. The sizes of the
rectangles are dependent on the number of child nodes to be partitioned.
The original layout of Treemaps is called Slice and Dice [9]. This
treemap layout is constructed by recursive subdivision of a parent node in
a vertical direction for one layer and a horizontal direction for the next
layer. The result of division reflects the hierarchal information of the tree.
The disadvantage of this initial layout algorithm is that the restriction of
division directions causes a density of rectangles in a high aspect ratio,
even for small data sets. The visual representation generated affects a
user's ability to recognize nodes, especially when data size increases. To
assess the effects of aspect ratios on readability, experimental evaluations
have been carried out. The results prove that extreme aspect ratios have a
significant negative impact on rectangular area judgments and
comparisons.
This problem has motivated many researchers to carry out research in
this direction. Recently, alternative Treemaps have been proposed,
including Clustered Treemaps [10], Cushion Treemaps [11], Squarified
Treemaps [12], and Ordered and Quantum Treemaps [13]. These
techniques typically satisfy one or more aesthetic criteria. To achieve
better results for symmetry and proximity, Clustered Treemaps optimize
neighbour similarities by re-arranging a set of “siblings” in the hierarchy.
To improve readability, researchers developed algorithms which attempted