International Semantic Web Conference

ISWC06, Athens Georgia, USA.

The proceedings are Springer LNCS 4273

Both Norman and Elizabeth were at this meeting. The following are Norman's notes on a few notable talks and conversations; Elizabeth's notes are at IswcNotes2006

Peter Fox at HAO

I had an excellent conversation with Peter, who was presenting some results of work on Semantically-Enabled Large-Scale Science Data Repositories, which is obviously relevant to us (his paper is at doi 10.1007/11926078_57, if you have access to that). He's done a lot in the STP and upper-atmosphere area.

Peter's closely involved with a project called OPeNDAP, which is a data abstraction environment which sounds more than pretty nifty. It sounds as if it encapsulates multiple data sources, and parts of data sources, at just the right level to have semantics added to them. And it can handle FITS.

STP is a bit different from night-time astronomy because Peter ends up being the curator for a lot of data, so he gets to talk directly to the data originators, and involve them. Also, the community seems happy to be given tools, so there ends up being a tighter cycle of the community being shown benefit from ontological additions.

They have a well-developed provenance effort.

Shenhui Wang -- analysing leaf-shape descriptions

A very interesting talk (doi 10.1007/11926078_48) discussing reasoning about the shapes of leaves: you can ask `I have a greenish-brown roughly elliptical leaf: what plant is it from?, and the system can turn `greenish-brown' and `roughly elliptical' into things precise enough to reason about. How is this relevant to us? Two reasons. First is that this sounds like the same sort of problem that we have when querying the registry. And secondly, this involves reasoning with numbers, and definining classes such as `all those with the value of parameter x in the range 10-20', which is not expressible in OWL-DL, but is in an extension of that called OWL-Eu, which she points to.

Their interest in this is in reasoning about natural-language descriptions, but I think it could work perfectly well as a species of fuzzy reasoning and querying.

Visualising RDF

There was least one whole session on this. Fresnel, /facet, IzViz were all described, and all are worth closer examination, I think.

Mulgara -- a scalable triple store and RDBMS SPARQL front-end

Mentioned Mulgara is a scalable Java open-source triplestore. There's a version of it which sits on top of an existing RDBMS and maps it to RDF on the fly. This is a reimplementation of the non-Jena part of Chris Bizer's D2RQ (which is reportedly very closely tied in to the underside of Jena and its BackingStore SPI). This is in SVN at present, but will be released before long.

Mulgara has a query language, ITQL, which is more expressive than SPARQL, has ontological support (RDFS now, OWL-Lite very soon; I think that most SPARQL endpoints can support this sort of thing by effectively creating implied triples, but it's not directly expressed in the language), and can do update/adding (neither of which SPARQL can do). There will soon be SPARQL support.

Mulgara has a query engine sitting on top of a `Resolver' SPI, which in turn sits on top of the triple store, breaks down a query into the multiple sets of matching triples, and then does a join to produce the query result.

Performance: The Mulgara triple store is scalable up to 0.5billion triples (with 64-bit Java). It's good for 100k triples and up, but overkill for 10k. The store is called XAStore, and XA2Store is coming soon, which has multiple optimisations and miscellaneous cleverness.

A Mulgara is an endangered Australian marsupial.

[ It seems there's a long story of the predecessors to Mulgara, open-source Tucana -> Kowari (more marsupials). The software, or the company (I forget which) was sold to Northrop-Grumman, who tried to close down the OSS effort with lawyers, which the original developer resisted with more lawyers, eventually getting a public apology from Northrop. Mulgara is a fork of Kowari/Tucana, branched just before Northrop put any code in. It seems that all the open-source development effort has gone in to the Mulgara fork, and the open-source developers have abandoned the Kowari/Tucana fork, and the two forks are becoming increasingly incompatible, and the Northrop-Grumman Tucana page has apparently been taken down. ]

Et cetera

There were a good number of other interesting talks (and some incredibly tedious ones -- there's only so many upside-down As you can cope with in one week), and probably a greater number of interesting conversations, which are less immediately transcribable.

-- NormanGray - 01 Dec 2006