Probing Projections: Interaction Techniques for Interpreting Arrangements and Errors of Dimensionality Reductions

Area of Science:

• Computer Science
• Data Visualization
• Human-Computer Interaction

Background:

• Dimensionality reduction techniques project high-dimensional data into lower dimensions for visualization.
• Interpreting these projections and understanding their inherent approximation errors remains a challenge.
• Existing research primarily focuses on projection generation and comparison, with less on interaction with the resulting visualizations.

Purpose of the Study:

• To introduce integrated interaction techniques for interpreting and interrogating dimensionality-reduced data.
• To address the difficulty in grasping the meaning and quality of data projections.
• To provide methods for examining approximation errors and the influence of data dimensions.

Main Methods:

• Developed the concept of 'probing' as an integrated approach to visualization interpretation.
• Proposed interactive methods to analyze dimensionality-reduced data and the projection itself.
• Created a web-based system to implement and evaluate these interactive methods.

Main Results:

• The proposed methods enable viewers to see approximation errors and question element positioning.
• Users can compare data elements and visualize the influence of specific dimensions on the projection.
• Evaluation with data analysts showed the utility of the prototype for examining multidimensional datasets.

Conclusions:

• Integrated interaction techniques, particularly 'probing', enhance the interpretation of dimensionality-reduced data visualizations.
• The developed methods improve understanding of projection quality and data relationships.
• The findings suggest a valuable new direction for interactive data visualization research and application.