EVM: Incorporating Model Checking into Exploratory Visual Analysis.

Area of Science:

• Computer Science
• Statistics
• Human-Computer Interaction

Background:

• Visual analytics (VA) tools facilitate rapid data exploration and pattern discovery.
• Existing VA tools often lack mechanisms for explicitly verifying data interpretations against underlying models.
• Analysts may struggle to validate if observed patterns align with hypothesized data-generating processes.

Purpose of the Study:

• To introduce EVM, a novel data exploration tool.
• To enable users to express and visually check provisional interpretations of data using statistical models.
• To evaluate the impact of model checks on data exploration processes in a user study.

Main Methods:

• Development of EVM, integrating statistical model expression and visualization-based model checks.
• Rendering distributions of model predictions alongside user-generated data views.
• Conducting a user study with data scientists to observe the use of model checks during exploration.

Main Results:

• EVM facilitates the expression and checking of data interpretations via statistical models.
• Visualization-based model checks help analysts scrutinize expectations about data-generating processes.
• User study participants actively used model checks to refine their understanding of data relationships.

Conclusions:

• Integrating model checking into VA tools enhances the rigor of data exploration.
• EVM demonstrates the utility of explicit model-based interpretation checks.
• Further development can scaffold model exploration within VA environments for improved analytical outcomes.