catchup thoughts

From: David R. Karger <karger_at_mit.edu>
Date: Mon, 6 Jun 2005 17:27:48 -0400

I've spent a whole week catching up on the fresnel list archives, so
now want to send out various thought. Probably too late to be useful,
but just in case, here is the first:

I am now of the impression that there are _three_ distinct layers in
our ontology. Right now what we call "selection" involves both
selection and grouping. We should separate them. The first,
selection, should make no assumptions about grouping. Chris has said
that this would leave selection as nothing more than a query, but I
disagree. What remains crucial is the decision of which selection
should be made at which time, i.e. by specifying the purpose that a
particular selection can accomplish.

It is this selection layer where I want to specify that "the summary
information about a person is their name and email address". More
generally I could imagine using the selection ontology to specify
which properties are "important", "advanced", "for debugging", "useful
for a quick browse", "keys (unique identifiers)", etc.

I'm not even sure we should bother to specify a particular selection
query language. If we do, I would focus on what useful syntactic
sugars might capture particularly common queries in our
application---for example, using just a list of properties to specify
that the values of all those properties should be returned, instead of
writing a corresponding sparql query.

Selection of what I want to show should produce rdf, not a tree. Some
devices might want to turn that rdf into hierarchical xml that can be
styled, but I should also be able to use the selectors to, e.g.,
decide what data to export into a database, in which case I want pure
rdf. Also, I probably want to use this selection language to produce
the RDF that gets fed to a graph-oriented viewing tool like isaviz.

I wonder if it would be useful to associate each property with a
"degree of detail" on some e.g. numeric scale, where 1 corresponds to
necessary in all circumstances and 10 to only being needed in
complete-info view?
Received on Mon Jun 06 2005 - 21:25:53 EDT