Raw

A quick post (ha, spoke too soon) for the benefit of the RDFers wondering about freebase/Metaweb. I scrounged an alpha account, and have emailed a bit with Robert Cook (parallel computing/db expert) & John Giannandrea (amongst other things worked with RV Guha on the initial RDF MCF submission [? see danbri's comment] to W3C while at Netscape). I'm not altogether sure what I can disclose, but I get the impression they aren't too worried, the closed-alpha is for bugfix purposes, they're aiming to go wide open before very long ( and pretty much all of this could be derived from Tim O'Reilly's post, if you looked sideways and squinted...). The following is just my opinion/bits I've gleaned, may not be very accurate. I won't even guess at their business model.

Ok, they've got a entity-relationship kind of graph model which is close to RDF, with a store that's currently approx equiv to a billion triples, and is capable of scaling up a lot more. They've preloaded it with a bunch of datasets, e.g. Wikipedia, MusikBrainz. The user interface is very impressive, the nearest analogy I can think of is if you imagine Longwell with editing facilities, Ajaxified big style.

They see the basis very much as a database, not an ontology kind of tool, although that's really a matter of definitions. (If pressed, personally I'd call Atom an ontology). Users can create sharable types & properties, effectively in their own namespaces. This is starting to make it sound a bit heavy-going, but it isn't at all. Tim O'Reilly wasn't wrong about calling it addictive (check his screenshots). I've only fiddled a bit, but it is satisfying being able to organise stuff with an intuitive user interface. Plenty of Creative Commons logos on display. What Google Base could have been...

I got the impression some of their input/output stuff is work-in-progress, final bits of front-end issues. One big thing they do have in place are easy JSON-based services. Basically graph pattern queries against the store, results in a similar shape. Yup, the QL seems conceptually very close to SPARQL, this could make a good starting point for a JSON-syntax version. They've obviously put a lot of effort into making it something your typical Web 2.0 developer can use within a few minutes of seeing. This URI appears to be publicly accessible: mjt ( I hope that's intentional) - it's a Javascript templating language they've put together, it's closely modelled on kid.

My main criticism to them has been that although the system is conceptually very close to standards like RDF, it differs to the extent that compatibility with other systems wouldn't be direct at this point in time. But I reckon that qualifier's deserved - they are aware of these standards, but have prioritised getting the system built & usable, which seems fair enough. The specifics of their internal model certainly don't matter - I bet half the triplestores on the planet have got a SQL DB in the backend. I may well have been overcritical. Although this thing is hosted monolithically, what they've got so far seems to be considerably more connected than e.g. Wikipedia. I get a feeling that they're champing at the bit, really keen to hook into other systems on the web, and I suspect they'll have a raft of standard WebArch & SemWeb friendly interfaces in place soon (hopefully in time for their Beta ;-) I also suspect they'll have some great 'lessons learned' kind of things to feed back into the semweb community. Whatever, they've already got a system that I reckon a lot of people are going to like & use .

~

PS. Semantic MediaWiki's Denny Vrandecic has also been kicking freebase's tyres - and suggests their business plan (!). Material's coming out thick and fast on the freebase blog - First Patterns: Data Modeling Idioms on 'Compound Value Types' (er, which appear to be N-ary Relations) and What is MQL? about their SPARQLesque JSON query language (afraid it doesn't say much, links to auth-required material).

Some comments below too - thanks! Credit for the use of em spacing in the CSS (apparently more popular than the content here ;-) goes to Reto & bengee. Bill, you're suggesting Atom isn't a "specification of a conceptualisation"?

 

@en