Is it possible to design an incremental RDF framework for such data ?
The main ingredients would be a starting "header" RDF block containing
feature declarations and metadata.
Followed by 1 incremental RDF "data block" at a time. Which would
essentially mean 1 row of the dataset at a time.
An optional end block could also be included to specify the stream end.

The "data blocks" could be very minimalist with just the uri
indentifier and the relevant data.

Its simply a matter of a special protocol  (within the RDF domain)
designed for the purpose of such large datasets.

RDF although large in their overall representation of objects provides
a lot of comprehension capabilities vs ARFF. Also theres the simple
matter of having a single extra  comma in your data crashing the whole
algorithm ( in ARFF) ! Natively RDF is more in tune to "comprehend"
data rather than represent data in an efficient form. Also have to
keep in mind that occasionally the need arises for a human to
comprehend such data (in the native machine form) and in such cases
RDF is quite unsuitable.

Another possibility could be to "compress" RDF using annotations. Of
course that would mean designing and standardizing the annotations for
all players who want to interact using the compressed form.

Just my two cents ....

Regds
Surajit

On 27 May 2011 21:20, chung <chvng at mail.ntua.gr> wrote:
> Hi All,
>    Some criticism on RDF from the experience we've gained in OpenTox :
> http://is.gd/qLJG3h . The article is not complete yet and will be
> enriched with more facts and diagrams. We also started an interesting
> discussion with Martin here : http://is.gd/cUKaCE which we can continue
> on the mailing list.
>
> Best regards,
> Pantelis
>
>
> _______________________________________________
> Development mailing list
> Development at opentox.org
> http://www.opentox.org/mailman/listinfo/development
>



-- 
Surajit Ray
Partner
www.rareindianart.com