Re: Owl and Piggybank

From: David Huynh <dfhuynh_at_csail.mit.edu>
Date: Wed, 23 Feb 2005 13:28:13 -0500

Hi Shashi,

Shashi Kant wrote:

>Hi David
>
>maybe this has been asked/answered before..apologies if these are newbie
>questions...but I had a few questions:
>
>1. If you are doing inferencing "locally" PB[Longwell] would that be
>isolated from other datasets? I guess my question is, do you see some form
>of distributed inferencing by allowing interactions across clients? Say my
>PB with yours..
>
>
Yes, it is done locally and per data import (i.e, on each chunk of data
you add, independently). This is not accurate because uniqueness of
property values should be detected globally. This extremely crude
approximation was made for the sake of performance. I am hoping that
someone else will come along and do a much better job than I did :-)
Hint, hint,...

>2. What kind of inferencing were you thinking of doing? I know this could be
>a laundry list running into thousands, but a few at the top of your list
>would help...
>
>
The inference I *had* to do was to support facetted browsing over
arbitrary data without explicit schemas. I need the following (and some
derivatives thereof):
- tell what the schemas there are/seem to be in the corpus, what classes
there are/seem to be in each schema, what properties each class
has/seems to have
- tell how unique values of a particular property are, whether they are
likely to be integers, dates, URLs, phone numbers, etc. (so that you can
facetted browse by area codes or domain names)

David
Received on Wed Feb 23 2005 - 18:29:47 EST