1. William Grosso Stanford Medical Informatics Stanford University
Aspects of Evolution
(Diagrams: Past, Present, and Future)
William Grosso
Stanford Medical Informatics
Stanford University

2. The Protégé User Interface
Classes are associated to forms
To acquire an instance, you simply fill out the form
To view an instance, you bring up the form

3. Slot Widgets are local mediators
Render a (frame, slot, value) tuple in a more amenable form
Allow editing and value manipulation
Collectively, the slot widgets form a translation layer
Where a widget is used is configurable
Widgets have restrictions (based on frame and value type)
Ontologists choose from the set of allowed widgets

4. Choosing Slot Widgets

5. Diagrams
Instead of viewing lots of forms, get a “high level” view of a related set of frames
Underlying the diagram is a set of frames in the knowledge base
Drilling down (double-clicking) on diagrammatic objects brings up the associated frame(s)

6. Protégé/Win Diagram

7. Protégé/Win Diagram Behavior
Widget is available to any slot that takes multiple instance values
Associates instances in the KB to each node in the diagram
Visual aspects are keyed on class types
Just as classes are mapped to forms (which can be laid out), users can choose shape and color of visual representation

8. Protégé/Win Diagram Problems
Single type of “wire”
Connection information not stored directly in kb (stored indirectly by filling out slots on node-frames)
Positional information not stored as frames
Stored outside the knowledge-base in non-frame format
Hard to incorporate for other uses

9. The Protégé-2000 Diagram

10. Summary of Changes Old problems fixed Palette added
Multiple types of connections (with color and endpoint customizations)
All positional information stored in kb
Not necessarily a good thing
Palette added
Defines “logical drawing area” (adjustable canvas, can be bigger than the screen)
Idea of “widget ontology” introduced

11. a higher-arity relation
Widget Ontology
Connector is really
a higher-arity relation

12. Nice Aspects We can store our diagrams out to OKBC servers !
Other groups can automatically generate the frames for a diagram
People doing “sketch-based” UI can generate diagrams which then get incorporated into the KB
We’re going to need a richer set of concepts to describe spatial locations for this to really work
Much more configurable than old version

13. Future Work: Extensions
Coping with a big logical canvas
Finding a node (or set of nodes)
Navigation around the diagram
Using thumbnails
Explicit support for the notion of sub-diagram
Selective display
“Display all nodes that are reachable in 3 or fewer connections [of types …] from [particular node]”

14. Future Work: Bayesian Networks
Bayesian networks are usually represented as nodes and arrows
Reified connectors make this easy to do in Protégé-2000 (at a simple level)
Acquiring probabilities will require TableWidget (not yet built)
The hard stuff:
Incorporating Koller’s Probabilistic OO notions
Incorporating validation and critiquing feedback within the diagram
Cool Problem

15. Future Work: Metrics
Protégé-2000 computes static metrics about a knowledge-base
Protégé-2000 records “knowledge-base edits” to log files
We need to come up with a set of abstractions for “knowledge-level” edits
Dragging a wire could, conceptually, be “connecting two objects”
But it’s 13 kb edits (counting positional information) !!
Cool Problem