- From: David Booth <david@dbooth.org>
- Date: Wed, 16 Jan 2013 21:13:53 -0500
- To: "Sivaram Arabandi, MD" <sivaram.arabandi@gmail.com>
- Cc: Peter.Hendler@kp.org, meadch@mail.nih.gov, conor-dowling@caregraf.com, d.rebholz.schuhmann@gmail.com, jluciano@gmail.com, michel.dumontier@gmail.com, public-semweb-lifesci@w3.org, ri@semanticidentity.com, rmrich5@gmail.com, tfmorris@gmail.com
Hi Siviram, On Wed, 2013-01-16 at 14:51 -0600, Sivaram Arabandi, MD wrote: > I am enjoying reading and catching up on this thread. > > David, you mentioned 'rdf model' below - are you referring to ontology > models? Yes, sort of. One can design an RDF model without formalizing it into a written ontology. So I was referring to the schema of the RDF data -- classes, relationships, etc -- whether or not that schema is implicit or explicit (i.e., written into an ontology). > And, you said "To my mind, monotonicity is the key." But in medicine > most reasoning is non-monotonic - default reasoning, (educated) > guesses and revision of diagnosis as new data comes into the picture. > What am I missing here? Today you might conclude "As of 16-Jan-2013 the diagnosis is X", but tomorrow you might conclude "As of 17-Jan-2013 the diagnosis is Y". If you represent the statements that way (qualified by the particular context or, in this case, date) then they are monotonic -- they remain true forever, regardless of what new information arrives. Whereas if today you were to represent that information simply as "The diagnosis is X" then it would be non-monotonic, because tomorrow you might need to change it to "The diagnosis is Y". OTOH, even if the data is monotonic, you can cleanly use default reasoning and the closed world assumption in the way you *use* that data. For example, if the data is represented like "As of 16-Jan-2013 the diagnosis is X", then a query can return the *latest* diagnosis and report that "the current diagnosis is X". Tomorrow, after more data has been added, that same query might report that "the current diagnosis is Y". You can think of "the current diagnosis is X" as being derived, non-monotonic data. If you store it, you must be careful to treat it only as cached information that will be invalidated when its antecedent information changes. The RDF Pipeline framework that I've been working on http://code.google.com/p/rdf-pipeline/ was designed in part to handle this kind of problem: to keep track of information dependencies and update derived non-monotonic information automatically. Here are slides about it from last year's Semantic Technology conference: http://dbooth.org/2011/pipeline/ You can also design your RDF data models so that certain pieces make closed world assumption or use defaulting conventions. Then you have to be careful to keep track of which pieces they are, so that when you merge data you do so appropriately without invalidating anything. For example, you might change defaults into explicit values before merging. -- David Booth, Ph.D. http://dbooth.org/ Loss of web prodigy Aaron Swartz: http://tinyurl.com/ahe2k8c Opinions expressed herein are those of the author and do not necessarily reflect those of his employer.
Received on Thursday, 17 January 2013 02:14:25 UTC