Document Visualization at UMBC

Document Visualization at UMBC

Penny Rheingans

October 2002

Goals

• Provide powerful and flexible visual representations of document data in order to facilitate exploration and explanation

• Develop novel information visualization and interaction techniques applicable to a wide range of document types

Document Metadata

• Date• Relevance

– multiple measures

• Topic– multiple keywords

• Provenance• Whatever

Information Visualization Toolkit

• Glyph-based visualization of document by metadata– Glyph per document– Metadata values mapped interactively to

• 3D position• Color• Size• Opacity

• Interactive 3D viewing• Built on vtk with tcl/tk interface

Recent Advances

• Implemented on both irix and linux platforms• Automatic and flexible scaling• Basic clustering

Ongoing Efforts

• Additional platforms• Performance enhancements• Picking and probing• Dynamic update• Closer integration with IR engines• Application to intrusion detection

New Directions for ivtk

• Additional visualization techniques, possibly– Additional dimension reduction methods– Parallel coordinates

• Improved user interface with expanded interaction modes (carried out as part of LUCITE task)

• Address issues of scale– Provide both context and detail on demand– Filtering of data items– Generalization and summary representations– Glyphs for composite objects– Hierarchical viewing modes (LUCITE)

Filtering of Data Items

• Select which individual items to show– Drastically reduce number of items displayed

– Subsetting mechanisms

• Interactive value-based selection

• Stochastic sampling

• Select how to show individual items– Direct attention to match interest

– Modulate appearance of items

Emphasis Filter Approach

• Treat match to target as variable• Map to perceptually-based parameter or alter base

mapping• Interaction capabilities

– dynamic control of view

– dynamic control of some visualization parameters

• Previous SFA-based prototype

Personal Profiles

• Individual weightings for documents– flexible

– audience-specific

– interactive

• Supplement to precomputed relevance– simple measures

– computed dynamically

Matching Mechanisms

• Value Requirement• Composite

– multiple factors– fuzzy matches

• Threshold– significance levels

• Window– focus area

Future Directions in Filtering

• Revive prototype• Extend to include profile mechanism

– profile construction and manipulation

– matching metric specification

• Integrate with ivtk

Summary Representations

• Convey overview of corpus structure without showing each individual document

• Possibilities– Visualization processes to create and render aggregates

• Splatting or other implicit representations

• Isosurfaces or volume rendering

– Explicit analysis to characterize structure

Glyphs for Composite Objects

• Extend glyph representation to apply to composite objects– Collections

– Clusters

• Information of interest about composites– Mean variable value

– Variable range/variability

– Number of component objects

Composite Metaphors

• Display mean and variability information in consistent manner– Position

– Color

• Some visual attributes dedicated to supporting other display parameters– Opacity

– Shape