AgentVis: Visual Analysis of Agent Behavior With Hierarchical Glyphs

Area of Science:

• Data visualization
• Information visualization
• Human-computer interaction

Background:

• Glyphs are effective for multivariate data visualization but suffer from overlap and occlusion.
• Visualizing large datasets, like thousands of call center agents, with traditional glyphs is challenging and does not scale.
• Existing methods limit the number of data points displayable in a single image.

Purpose of the Study:

• To develop a scalable glyph visualization technique for large datasets.
• To address the limitations of overlapping and occluded glyphs in multivariate data representation.
• To improve the visualization of complex agent behavior in high-density environments like call centers.

Main Methods:

• Developed a hierarchical glyph clustering approach to group overlapping glyphs into a single parent glyph.
• Explored and implemented multivariate clustering techniques in collaboration with call center industry experts.
• Implemented dynamic control of glyph clusters based on zoom level and customized distance metrics.

Main Results:

• The proposed method effectively reduces visual clutter and overplotting by clustering glyphs.
• Hierarchical glyphs represent the mean values of clustered data, simplifying complex visualizations.
• The technique successfully visualizes thousands of call center agents, revealing behavioral insights.

Conclusions:

• Glyph clustering offers a scalable solution for visualizing large, complex datasets.
• The developed technique enhances data analysis by reducing visual clutter and improving clarity.
• This approach provides valuable insights into agent behavior, validated by industry experts.