My PhD dissertation and supporting documents can be obtained from the Virginia Tech library. In this work, I deeply explored the simultaneous application of dimensionality reduction and clustering algorithms in interactive visual analytics. In exploring the existing literature, I looked at facets of both the visualization of these algorithms as well as interactions with these algorithms. I developed a framework for structuring tools that use these algorithms, and I implemented three prototype tools to demonstrate trade-offs between these approaches (named Castor, Pollux, and Gemini to bring in my astronomy hobby). The dissertation concludes with a study exploring how users would like to see data structured with these algorithms and a research agenda for human-in-the-loop applications more broadly.
The overall document has the following chapter structure:
- Introduction
- Background: Dimension Reduction Algorithms and Tools
- Background: Clustering Algorithms and Tools
- Dimension Reduction and Clustering Projections
- Dimension Reduction and Clustering Interactions
- Castor: Dimension Reduction First
- Pollux: Clustering First
- Analyzing Pipeline Order Via Case Studies
- Cognitive Dimension Reduction and Clustering
- Human in the Loop Research Agenda
- Discussion and COnclusion
