Raw

After roughing out the conversion of Open Social People data into RDF, I felt obliged to have a look at going the other way (not least because a ConverterFromRdf can be as handy as a ConverterToRdf). Not quite there yet, but can't really get any further without being a bit more clued in about what's in FOAF in the wild, and will actually be done with the stuff. 

I'd got most of the necessary code from when I played with FOAF to hCard via SPARQL. That was quite a while go, so I decided to go over that again to get fresher data. To start I seeded a quick & dirty scutter with the profile list I got off the FOAFBulletinBoard last time. I did make one change - added timbl's FOAF, which he's been actively maintaining and is probably the richest I've seen. Silly me, I should have added GRDDL capability to the scutter (should have been trivial with the Jena GRDDL Reader), could have slurped XFN and hCard data, at least where people had bothered to use profiles.

I forget to note how many statements were aggregated this time (going 2 steps), but I think it was quite a few more than last time, around about 20k. Once again I had to manually delete a couple of statements with bad URIs which wouldn't serialise to RDF/XML (paypal, for foaf:tipjar). Here's all.n3.gz (180kB) and all.rdf.gz (245kB) I'm still not certain of what best to put in the SPARQL query for OpenSocial, but it'll probably be very similar to get-people.rq. This is a bit looser than the query I used to get hCard data last, and after running the results through sparql2hcard.xsl produces quite a lot of hCards. One thing I'd totally forgotten is smushing - there are quite a few near-duplicates. 

Anyhow I've now got most of the triples in my Talis Platform store, so as and when I try to figure out what exactly will be needed for OpenSocial I can experiment over a useful-sized dataset - here's an interface to the SPARQL endpoint. (Only 'most' of the triple because I wound up splitting the n3 in two, before converting it to rdf/xml for upload - I got proxy errors with the full 1MB version - maybe related to my slow line).