Re: CSV2RDF redraft

On 26/03/14 17:44, Ivan Herman wrote:
> I think one thing is to have a header (or not) and the other is to
> have a metadata.

Agreed.  I'm not sure what Jeni is proposing but without at least one of 
headers or annotations, it's data exchange (knowledge agreed between 
parties out-of-band, not accessible to anyone else), not publishing.

That's not an RDF specific comment; it's true for JSON and XML.  All you 
get is a data structure and you might as well use the CSV file directly, 
parsed with your choice of CSV parser.

> In other word, it would be possible to have a data set without any
> header, but with extra information in the metadata file. So the
> question may be: what is the minimum information that a metadata
> should have to make any sort of a meaningful conversion?

Good question.

> I guess the
> answer is quite easy, but it also shows the fact that there may be a
> very close relationship between the metadata and the conversion
>
> (Which puts CSV into a different situation than RDB's, hence we have
> to be careful in taking the RDB2RDF results too blindly.)

I see it as "informing", not prescriptive.  If nothing else, a CSV file, 
with headers, still has much less information to work with.

One CSV file is more like a bunch of denormalized tables.  We will 
succeed as a WG without needing to renormalize.

I hope that is clear by example in

https://www.w3.org/2013/csvw/wiki/CSV2RDF

As for groups of files, I think the WG-minimum is see that as a loose 
association.  Link generation that happens then to connect files is much 
easier than defining a process that converts files in a group with data 
flowing between files.

	Andy

>
> Ivan

Received on Wednesday, 26 March 2014 18:48:01 UTC