Re: [Tracker] tracker RDF data

[Eyal Oren <eyal oren deri org>]

On 07/19/06/07/06 15:54 +0100, Jamie McCracken wrote:
> Eyal Oren wrote:
> tracker is not designed for use as a web service but as a desktop service 
> (hence dbus is used).
I know, that's good. RDF is definitely not only for the web.

> Its also not an RDF style application it merely uses RDFQ (xml based 
> version) as a means to query stuff. We dont implement all of RDFQ but a 
> subset with extensions. We also only uses the <rdfq:Condition> parts in 
> the query so its not a full blown RDF thingy. (you can see rdf query 
> examples in the tarball in the rdf-query-examples folder)
ok. but your datamodel is exactly like RDF: desktop-object has-property 
with-value, e.g. (file-abc creation-date 2006/01/01), or (contact hasName 
'jamie') seems exactly the same as the RDF model (subject has-predicate 
with-value), which is not so strange, because RDF was designed to be a 
simple and generic datamodel and directed graphs are then a natural 
solution.

I don't mean to say that you should internally use RDF, since e.g. MySQL is 
certainly more developed than current RDF databases. But what would be 
useful is to align terminology, as you do in the 'shared file metadata 
spec'. As you state there, the only requirement is that metadata names are 
unique and not overloaded: 

If I say 'file-abc is-owned-by EyalOren' and you say 'file-abc has-owner 
EyalOren' then we have a problem merging and combining those two pieces of 
metadata, and that's why you want to standardise the naming of these 
metadata indicators (called properties in RDF).

The thing is that in RDF a lot of work has been done into standardising 
exactly such vocubalaries. Such schemas could possibly contain indications 
on domain/range of those properties, e.g. has-creator usually relates files 
to persons.  See for example schemaweb [1] for some existing schemas (the 
PIMO schema [2] used by gnowsis is very relevant, but the document is very 
long and winded.

If you are interested, I would certainly be interested to figure out how we 
can better align the existing schemas in the RDF world with your 'shared 
file metadata specification'.

Of course, if you're not interested in that, I can easily do the 
tracker-terminology-to-rdf-schema-vocabulary translation myself, in the 
adapter (as we did for evolution data: we mapped it to FOAF [3] and some 
other RDF schemas).

> The dbus stuff will need Ruby bindings (if they exist?). Queries are 
> returned via dbus as a native dict/Hashtable  with the URI as the key and 
> an array/list of values as the contents. I expect if you were going to 
> make use of it you could serialise this into ntriples yourself.
Yes, the triples are no problem, but the ruby dbus bindings are very old 
and seem to be deserted. I don't really know how to get that working, I 
don't know anything about dbus myself, so I don't see myself writing those 
bindings.

> Im not sure if the above helps or not but it sounds like it would need 
> some mods on your part to make use of it.
Yes definitely. If I can get the dbus stuff to work somehow, I can easily 
do the rest. 

What kind of queries can I send to org.freedesktop.Tracker.Search.Query?  
Which subset of RDFQ do you support exactly?

-eyal

[1] http://www.schemaweb.info
[2] 
http://www.dfki.uni-kl.de/~sauermann/2006/01-pimo-report/pimOntologyLanguageReport.html
[3] http://www.foaf-project.org/