More Like Vis, Less Like Vis: Comparing Interactions for Integrating User Preferences Into Partial Specification...

Area of Science:

• Human-Computer Interaction
• Information Visualization
• Data Science

Background:

• Visualization recommendation systems automate data exploration, aiding non-expert users.
• User preferences guide these systems via partial specifications.
• Interaction approaches for partial specifications remain underexplored.

Purpose of the Study:

• Compare three interaction approaches for partial specification input in visualization recommendation systems.
• Evaluate trade-offs between manual input, whole-visualization ratings, and component ratings.
• Investigate impact on data exploration, visualization diversity, and user comprehension.

Main Methods:

• A between-subjects study comparing three conditions: manual input, whole-visualization like/dislike ratings, and component-level like/dislike ratings.
• Participants completed a data exploration task using a visualization recommendation system.
• Qualitative feedback was collected to understand user strategies and comprehension.

Main Results:

• Manual input resulted in greater coverage of data dimensions.
• Like/dislike ratings (both whole and component) led to increased diversity in marks and channels used.
• Qualitative data revealed distinct user strategies and comprehension levels across conditions.

Conclusions:

• Different interaction approaches offer distinct benefits for visualization recommendation systems.
• Manual input is effective for comprehensive data exploration.
• Rating-based approaches enhance visualization variety and may impact user strategy.
• Findings inform the design of more effective interactive visualization tools.