Ontologies

• Task oriented. Start by focusing on resource discovery rather than, eg data modelling.
• Find out what technologies exist. CDS already doing some of this - talking to IRIT in Toulouse. Extract from papers and relate to IAU Thesaurus -> document based descriptions. --> High level ontology, run it against keywords in the registry resources. Could immediately implement (information already in resources, ontology exists at IRIT). Need a lower level ontology for things like date of observations/bandwidths.
• Investigate existing ontology/neural-net/dictionaries/etc?
• UCDs are a set of terms that are a taxonomy (a heirarchical relationship of terms); ontologies are things that relate taxonomies to each other.
• eg, with an ontology, you can ask 'give me white dwarves' and you'll also find things that don't mention white dwarves but do have appropriate luminosity/temperature ranges, because the ontology says white dwarves 'have a' certain temperature ranges.
• What Registry-investigating ontology-building tools can we add to read through Resources and create registry ontologies.
• factguru.com has a ready made ontology

Data Models

Anita showed the progress on the IVO observation data model and how it is used to describe data that might be typically part of an investigation.

After some irrelevent discussion on data modelling, we realised this group won't design/build data models, but instead will make use of them as they arrive from other IVO organisations to structure resource discovery queries/investigations into data.

Discoverers:

• Humans - need pleasant guiding UI
• Agents/other software - need language to describe task.

Add some kind of brokering agent that can look first in the registry but also go off around the services asking low level information to do with resolution, keywords (with ontology links).

Resource Investigation Language

Looks like we need a special language for describing (eg to agents/discovering broker) not just what we're looking for but how it is constrained. Combine with ontologies in order to relate terms together. Link with (possibly) VOQL / agent negotiation languages / needs prototyping etc.

Need to be able to specify 'tightness' - how good enough must the returned list of resources be - as well as a description in the response about decisions made (eg, the broker looked not just for 'young stars' but these other related terms)

Similarly need to specify how deep the broker is to look; eg stop at registry, or look deeper, or look for general high level concepts (spiral galaxies) or low level (limiting flux/location/observation times/etc).

Prototypes

Possible prototype: simple use case (Sebastian's late type star -> all 'subclass' of late type stars, or if 'TTauri star' search returns null, broaden to 'superclass' young stars.

One-step computed terms

For example, a search for 'nearby stars' might also find terms that say that parallex is a measure of distance and 'nearby' is less than, say, 100 lys, so the investigator can do an extra query on this computation.

Workflow-building investigators

For example, a search for 'AGNs' might include the broker, if it knows the characteristics of the AGN (through the ontology) and what tools can process the appropriate data appropriately, then it can build a workflow, find the candidates, and return those, as well as just returning the resources with the appropriate keywords. For fun investigation...

Improving Registry Resources

What needs to be added to the Registry Resources to support investigating feet.

-- MartinHill - 19 Nov 2004 - apologies for spelling, smelling, abuse, etc