W3C home > Mailing lists > Public > www-rdf-interest@w3.org > October 2004

pointer: MSGs as alternative (extended) CBR

From: Giovanni Tummarello <tummarel@ascu.unian.it>
Date: Fri, 1 Oct 2004 19:21:52 +0200 (CEST)
To: www-rdf-interest@w3.org
Message-ID: <Pine.LNX.4.61.0410011907450.17502@ascu.unian.it>

Hello all, i apologize for not quoting the perfect message here and for 
being expecially terse, i am out at a conference and dont have the time 
for a full analysis of the very rich thread.

I just want to point out our experience and implementation
In the  DBin (www.dbin.org) semantic web P2P system each node acts as 
information source about all the URI tha belong to a certain class (of 
interest.. say "musicians and garlic spiced food"). Being so, CBR seemed a 
good start.. but the asymmetry in our opinion wasnt enough .. or 
at least didnt fit our general view. There is nothing that say that an 
ontology ahs tobe active or passive in its terms. To cope with this we 
"minted" the term "relates" and say that a statement a relates to be if a 
or b are either subject or object respectively. From this we work a 
definition that is similar to the CBR . We callthis MSG minimum self 
contained graph. Reason for this is that it can be proven that that graph 
has sense in itself, and can recostruct a whole model on another peer 
(transferring 1 msg at a time).

There are other nice properties (given a statement its uniquely clear 
which msg it belongs to etc..) that lead to intesting properties about 
trust and provenience of iformation .. e.g. you only have to reify 1 
statement in a MSG  to be sure that the whole MSG comes from a specific 
source. MSGs also have a fixed or linera with the number of blank 
noddes execution times and that forms the base for the scalability of 
the whole DBin project as opposed to other approaches which spread (and 
demand the execution) of arbitrary queries.
Its all detailed in the paper at:

http://www.dbin.org/twiki/pub/About/WebHome/RDFGROWth_workshopISWC2004.pdf

the paper is going to be presented at iswc along with dbin itself (there 
will be a separate announchment about this on this list once we get a 
better version out.. should be days)

About the original CBR.. i believe it also has the property that you can 
transfer a whole graph just by adding cbr and not necessairly MSGs.. 
however MSGs (or some reviewed version to comprise OWL IFP) perfectly fit 
a neutral "semantic web server" approach.

Once again here.. i totally agree with the general view of Sticker here. 
These tools fit some task .. some useful task. If one thing is fairly 
useful i will be interested into it and promote it even if its not 100% 
theoretically perfect. its the most 
promising part of the semantic web :-)

Giovanni
Received on Friday, 1 October 2004 17:25:57 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 December 2009 10:52:09 GMT