Podium: Ranking Data Using Mixed-Initiative Visual Analytics.

Area of Science:

• Data Visualization
• Human-Computer Interaction
• Machine Learning

Background:

• Multi-attribute ranking systems are common for data-driven decisions but require users to quantify attribute importance, which is often difficult.
• Users typically possess a holistic understanding of data rather than precise attribute weights.

Purpose of the Study:

• To present a visual analytic application, Podium, that assists users in ranking multivariate data points.
• To enable users to infer attribute importance from their relative data preferences.

Main Methods:

• Developed a prototype system, Podium, allowing users to rank data by direct manipulation (dragging rows).
• Employed Ranking Support Vector Machines (Ranking SVM) to infer a weighting model from user-defined rankings.

Main Results:

• Podium infers attribute weights that align with user preferences, facilitating understanding of subjective data value.
• The system aids in identifying attributes that influence user perception and deconstructing importance in existing rankings.

Conclusions:

• The Podium application democratizes the use of machine learning for data ranking and preference analysis.
• This approach enhances user comprehension of data by revealing underlying attribute significance without requiring explicit quantification.