Overlap Removal by Stochastic Gradient Descent With(out) Shape Awareness.

Area of Science:

• Computer Science
• Data Visualization
• Computational Geometry

Background:

• Data points in 2D visualizations are often represented as shapes, but inappropriate size/shape choices can cause overlaps, obscuring information.
• Overlap Removal (OR) algorithms are post-processing solutions to ensure graphical elements accurately represent underlying data.
• Preserving the original data layout's topology is crucial for effective visualization exploration.

Purpose of the Study:

• To extend the FORBID algorithm for Overlap Removal (OR) in 2D visualizations.
• To develop a shape-aware OR algorithm (SORDID) capable of handling polygonal shapes.
• To achieve an overlap-free layout balancing compactness and original layout preservation.

Main Methods:

• Modeling OR as a joint stress and scaling optimization problem using stochastic gradient descent.
• Developing SORDID, a shape-aware adaptation of the FORBID algorithm for polygonal data points.
• Comparing proposed approaches against state-of-the-art algorithms using quality metrics.

Main Results:

• The extended FORBID and SORDID algorithms effectively remove overlaps in 2D visualizations.
• The methods achieve a balance between data compactness and preservation of original layout structures.
• Evaluations demonstrate superior performance compared to existing state-of-the-art OR algorithms.

Conclusions:

• The proposed OR algorithms offer significant improvements in visualization clarity and data exploration.
• SORDID provides a robust solution for Overlap Removal with arbitrary polygonal shapes.
• These advancements enhance the reliability and interpretability of complex 2D data visualizations.