dynamic data and actor views using ontologies
Current lists of actors (and planned data sets) are static in that the tree is
statically written into a MoML model and displayed. This severely limits the
user's ability to find appropriate actors (and data sets) as the number of
actors grows. The current tree is a combination of functional and project
oriented folders, with no consistent classification.
This proposal is to generate dynamic views of the actors and data sets by
organizing the actors into trees using simple ontologies and controlled
vocabularies. Each actor (in its MoML code) and each data set (in its metadata
description) would contain term references that are drawn from one or more
ontologies. For example, an actor might be classified as belonging to the Class
"SimulationModel" while another actor might belong to the Class
"AnalyticalModel". If both AnalyticalModel and SimulationModel are subclasses
of "Model", then we could display a dynamically generated tree like this:
Model |__ SimulationModel |__ AnalyticalModel |__ NumericalModel |__ IndividualBasedModel
with each of the Actors displayed at the appropriate node in the tree. Of
course, if SimulationModel has subclasses itself, those could either be
collapsed to show all models under SimulationModel, or additional levels of the
tree can be added.
The same scenario applies to data sets, allowing people to browse data according
to a particular classification ontology. For example, data could be classified
as applying to certain types of measurements:
BiologicalMeasurement |__ MolecularMeasurement |__ CellularMeasurement |__ TissueMeasurement |__ OrganismMeasurement |__ PopulationMeasurement |__ CommunityMeasurement |__ EcosystemMeasurement
Although this example is somewhat contrived, it illustrates the type of ontology
one might use. Need to talk to some domain scientists to determine an
appropriate set of classifications for data.
Switching classification schemes would be done dynamically, on-the-fly. The set
of ontologies that are available for display would need to somehow be limited to
a meaningful set (all of the classes in even a small, simple ontology would
overwhelm the user). This could probably be set through a configuration. In
addition, the ontologies would need to be stored in a Kepler-accessible
location, possibly included with the release.
#1 Updated by Chad Berkley about 16 years ago
Another major issue with this is how to tag atomic actors that have no xml
description. One option is to have some sort of LUT, possibly in a web service,
for linking components to their ontological description. also need to figure
out how to add this information to existing moml files without upsetting the
#3 Updated by Shawn Bowers about 16 years ago
We need to clarify the "query language" that will be provided to allow a user to
dynamically view actors and datasets via ontological information.
The simplest "query language" would be to allow a user to select a single
concept to organize the view by. For example, by giving "AnalyticalModel", a
tree would be created with "AnalyticalModel" as the root node, the various
subtypes of "AnalyticalModel" as internal nodes, and the actual actors/workflows
as leaves of the tree. (Note that we prune internal nodes here if no
actors/workflows exist along the path.)
As more complex "query expressions" are permitted, how results should be
displayed becomes less clear. For example, we could permit single concept query
plus role restrictions such as "AnalyticalModel and uses StatisticalModel". In
this case, it isn't clear how the root node should be displayed, or how further
subtypes of the root are best displayed. We could permit multiple concepts in
query expressions (implying that each concept is "anded"). Multiple concepts
with role restrictions. And, arbitrary concept expressions (e.g.,
"AnalyticalModel using StatisticalModel or StatisticalRegressionModel"). Each of
these would seem to complicate the display.
As a first cut, I think we should stick to single-concept queries.
#6 Updated by Chad Berkley almost 16 years ago
This is now working via Jena. Shawn and I have changed the kepler configuration
to only include a long moml list of actors and we have added an id property
which uses lsids for unique identification. Shawns code then takes the list of
actors and an ontology written in OWL and rebuilds the actor library based on
the ontology. This should be working by the alpha3 release.