From the problem discussed in the abstract, this study aimed to obtain better visible performance of small elements as well as achieve dynamically stable results, resulting in the input dataset being visualized in a relatively stable position of the treemap. However, satisfactory static and dynamic performance can be difficult to achieve at the same time in one treemapping method and it is for this reason that the study put these two directions into one framework. In the framework, we have three sub-branches: one has good static performance while with worse dynamic performance; one has good dynamic stable results while hard to visualize all small elements; one has static and stable dynamical performance between the former two.
This study proposed effective treemapping strategies that either maintain near-square aspect ratios for the rectangles of a treemap so that small elements are more visible or create stable layouts for time-dependent sequences (stock market variation with time), while these two aspects can be weighted in their influence. Three alternatives for arranging the results are thus designed: a size-balanced visualization maintains nearsquare aspect ratios, and a number-balanced version preserves the temporal coherence. An in-between is formed by a sequence-balanced partitioning method that maintains both aspects reasonably well.
For better comparison and evaluation, the study introduced a new metric that is defined based on the visual changes. It complements existing measurements for dynamic stability that usually only consider distance and size changes. This metric reflects shape variations, in addition to absolute distance and size changes. By comparing and discussing various methods using this measure in addition to other established measures, the study presented the behavior of the treemapping results with respect to standard datasets. 、
In the future, it is recommended that studies continue minimizing variances and maintaining temporal coherence. The use of a non-sliced method to divide rectangles is also an interesting exploration direction.
For the number-balanced partitioning method, we can combine hierarchical clustering and layout methods to not just two as close as possible. Moreover, for digit twins, the methods proposed in this study can be used for hierarchical dataset layout and visualization, which can be helpful for digit twin pipelines.
Virtual Reality & Intelligent Hardware
A Balanced-Partitioning Treemapping Method for Digital Hierarchical Dataset
9-Sep-2022