[ckan-discuss] package relationships

[David Read]

On 16 February 2011 23:55, David Raznick <kindly at gmail.com> wrote:
> On Wed, Feb 16, 2011 at 10:40 PM, Rufus Pollock <rufus.pollock at okfn.org>
> wrote:
>>
>> On 14 February 2011 18:33, Stefano Costa <stefano.costa at okfn.org> wrote:
>> > Hi CKANers,
>> > I'd like to ask (again) guidance on package relationships. This is a
>> > much wanted feature in the Italian case, because of the large number of
>> > datasets that we will be able to get by means of webscraping,
>> > translating to sane formats and other ways. We would like to be able to
>> > track derivative packages. In fact I've already created such a
>> > relationship http://it.ckan.net/package/linked-open-camera as a test.
>> >
>> > What I'd like to know is whether this way of creating relationships is
>> > going to become the standard way or something different (e.g. based on
>> > extra key-value pairs) is planned/proposed.
>>
>> I think proper relationships are probably best though one can abuse
>> key/values for this :)
>>
>> The main thing missing (at least from linked data perspective --
>> correct me if i am wrong Richard!) is the ability to add more info to
>> the link -- i.e. want not just to link but to say there these are
>> linked and have 100k triples pointing between them.
>>
>> Adding additional
>> metadata to links is reasonably straightforward but does require a
>> change to the core model and hence a migration. If someone wants to
>> jump in and try this they'll get full support and guidance from the
>> devs :)
>
>
> Package relationships already have a 'comments' field and this is used in
> the current ckan.net relationships.  So this is already implemented :)

Absolutely, but I think Rufus means having structured fields - extras.
His example, describing how many triples link between the datasets, is
currently put into the package extra fields (rather than the
relationship comment) because having it structured makes it easier to
collect in a script. But really this metadata is more associated with
the relationship than the package, so we should store the info there.

> The comments are in the bracket next to the link in the wui.
> However, the people involved with this still ended up using packages extras
> anyway with a link:package=value convention.
> For linked packages should we follow their convention?   If we do, we should
> make anything of the form link:package a link in the same way trac does.
> This would be nice as you could add links in the description field of the
> package too.
> Nonetheless, I personally agree that package relationships better suit
> *linked datasets*.  I do not think however that package relationships suit
> *derived datasets* better than the resource group proposal.  Currently the
> derived relationship type is not being used in ckan.net at all.

As has been said before, there's no edit UI for this feature yet -
it's yet to be completed - so it's not surprising no-one's using them
in this way yet.

I agree that Resource Groups are a better way to represent derived
data - where the metadata is similar e.g. that which has been cleaned
up. But relationships may be more useful where the metadata is
somewhat different - e.g. a combination of two other package datasets.
I think we're happy to give people reasonable options and watch which
ones take off.

David

> David
> _______________________________________________
> ckan-discuss mailing list
> ckan-discuss at lists.okfn.org
> http://lists.okfn.org/mailman/listinfo/ckan-discuss
>
>