International Semantic Web Conference (ISWC) 2006 DAY 2

Wednesday, 8 November 2006

Susie Stephens of Oracle presented her Industry Track paper Integrating Enterprise Data with Semantic Technologies. Oracle 10g Release 2 embeds a SPARQL-like graph pattern into SQL (that is industry speak for "We don't support SPARQL, but please don't penalize us for it"). There is apparently no funded project within Oracle to support SPARQL. A forward chaining rules engine, equivalent to Mulgara's Krule, support RDFS and user-defined rules.

She said that Oracle has put up to 1 billion RDF statements into Oracle. That tells me that we had better get onto funding Mulgara's XA2 next-generation data store in order to stay relevant in terms of scaling. However, she only showed numbers for query speeds up to 80 million triples.

Oracle's advantage is building on their mature database, which allows them access to existing features, such as encryption or scaling or clustering. A nice example of combining previous and new features was shown which combined a multimedia search with a term equivalence being given in RDF. Thus, a search for X-rays of "jaw" could pick up those tagged with the term "mandible".

Oracle 11g will support some level of OWL, reportedly something "similar" to OWL Lite but modified to reduce the ability to perform computationally intensive searches.

Oracle seems to have made it easy to get started with Semantic Web technologies. That is a good and positive thing for everyone in the industry. Her comments regarding modifications to the existing URI-based identifiers to allow use of existing unique identification schemes concerned me, though. Semantic Web technologies without URIs would be a huge step backward, even taking into account the obvious short term gains. Better to facilitate the mapping of existing identifiers to URIs.

Susie also mentioned that webMethods has announced RDF and OWL support in the new version of Fabric. Indeed, this press release from webMethods says that Fabric uses RDF and OWL. Specifically, "the library automatically learns dependencies and relationships between IT assets."

Explaining Conclusions from Diverse Knowledge Sources (J William Murdock, Deborah McGuinness, Paulo Pinheiro da Silva, Chris Welty, David Ferrucci). Their Open Source framework, Unstructured Information Management Architecture in Java, seems worth a look. The goal is to produce good search results when dealing with a mixture of structured and unstructured content. The example in the talk involved some textual data extraction coupled with some theorem proving.

There was an active Semantic Web Services track at the conference. This is hardly surprising, since UDDI is so badly broken and Web Services are left without a reasonable way to perform composition and discovery. Yet Semantic Web Services still seem to be mired in academia. Perhaps the industry will start to see the light if Oracle and webMethods successfully deploy useful semantic tools to the community.

A Software Engineering Approach to Design and Development of Semantic Web Service Applications (Marco Brambilla, Irene Celino, Stefano Ceri, Dario Cerizza, Emanuele Della Valle, Federico Michele Facca) described a top-down approach toward annotating Semantic Web Services using a Spiral development model. This was the first time I have seen anyone actually use WebML. I am going to have to look at that, especially since there is a tool which implements it. They also used WSML. The link provides an interesting summary of Web rules languages.

RS2D: Fast Adaptive Search for Semantic Web Services in Unstructured P2P Networks (Matthias Klusch, Ulrich Basters) presented a model for Open World searching of semantic services. They introduced concepts like Semantic Gain, Semantic Loss and a Baysian-derived risk factor to judge the likelihood that peers would have something to add to an answer. The idea is to use machine learning to reduce gratuitous network communication when querying Semantic Web Services. The algorithm works well for unstructured, peer-to-peer networks without a single authoritative source for information. This was also a best paper nominee.

Web 2.0 panel: Tom Gruber of realtravel identified "Collective Intelligence" as the critical feature of Web 2.0 and the area where the Semantic Web can provide the most value. Truth and semistructured queries are the critical components. "Don't ask what the Web knows, ask what the World knows", by which he means, "ask what people know."

The problem with SemWeb apps seems to me to be that they are almost all closed world. We need some large-scale, open world applications. Simile or Tabulator are the closest I have seen and incredibly cool, but they are far from mainstream. We need to encourage the development of more SemWeb apps which pull data from the Web and publish data back to the Web. I have been as guilty as anyone else on this, but I promise to try to get better.

Tom Gruber suggests that any app to address this problem should explicitly allow others to mash up on top of it. That is an excellent point.

Patrick Stickler of Nokia and Marja-Riitta Koivunen of Annotea and I discussed the state of SPARQL at dinner. The lack of a simple syntactical means of performing negation is a real problem. Perhaps we can fix that before SPARQL becomes a W3C Recommendation. Perhaps too I should recover my notes on Mulgara/Kowari/TKS's EXCLUDE operator and it's relation to Jena's NOT. We don't need an RDF query language standard that is hard to use, hard to implement and has two different types of null...