Part of the trouble with ontology/vocabulary discussions is that they
can fly away so easily into abstraction. I would like to try and ground
the discussion with some examples that I have heard over the last few
days. How can we make the UCD set so that the following are possible?
(1) Selection of events
A client may wish to get, for example:
"All events that involve cepheid sources" "All supernova events brighter than 18 magnitude" In order to service these subscriptions, the client would use specific UCDs in place of the natural language words "cepheid" and "supernova". We assume that a machine-learning system in the data pipeline has made probabilistic assessments that the event is "cepheid" or "supernova". We could also do this without the formal UCD structure by doing keyword search in the natural language "description" section of the event -- which of course assumes that some human had written the textual description.
(2) When a followup changes the hypothesis
Suppose there is an interesting event that is classified somehow (eg
supernova). There might be many followup observations that cite the
original event ID, and some of these might show a different
classification. How can we decide if the classification status of an
event has significantly changed, or become more definite, due to followup?
Received on 2005-04-27Z16:53:53