Ivan’s private site

October 28, 2009

ISWC2009 2-3

Second day

In fact, there is much less to say… In the morning I was on two workshops; I was at the Uncertainty Reasoning on the SW one for a while, but then I was asked to participate at a panel at the Semantics for the Rest of Us one, so I had to switch. This was a bit unfortunate, because I could not really ‘dive in’ to any of the two. And my afternoon was taken up by ‘networking’, catching up with some people on many many issues that are not worth blogging (yet?).

I listened to Kathryn Laskey’s presentation on how to combine probability theory in the mathematical sense (the good old Kolmogorov axiomatic theory on probability that I learned at university in a distant past…) with first order logic. I cannot claim to have really understood all the details but it made me curious enough to put reading her paper on my to do list…

As for the panel “Little vs Large Semantics: What’s next for the Semantic Web languages?”, with Leigh, Kendall and Ora on the panel besides me… it was not that exciting, I must admit. Maybe the main message I take away from it was the passionate request of Chris Welty to re-open RDF (see also Pat’s keynote below!).

Third day (well, first real conference day)

Preamble: I would have wanted to add links to papers. And I couldn’t: I have not found the papers on the Web. Neither on Springer’s site nor elsewhere. I may have missed a reference somewhere, if somebody knows then tell me. But if the papers are not available, I think it is a shame…

The conference began with a keynote of Pat Hayes. Entertaining and also thought provoking; Pat is a great speaker. What really interested me is his talk on ‘RDF Redux’; I was actually anxious to listen to that one at SemTech last June but he had to call this off back then. So he repeated it here. This is typically the kind of talk that needs more thinking afterwards to understand it (and Pat has promised to write it down!), but he essentially proposed to re-think and re-do some of the fundamentals of RDF semantics. Instead of set-based model theory which we have today, and which makes the treatment of b-nodes, shall we say, a bit complicated (some would use harsher words:-) we should consider RDF graphs as ‘things’ on a ‘surface’ (think of it as a real surface on a sheet of paper) and b-nodes are just ‘scratches on that  surface’. (A bit like ‘context’ of a graph?) Because these surfaces are different from one graph to the other, when a merge occurs then in fact a new surface is created where the unified graph is put, and the issue of b-nodes becomes natural (instead of the ‘renaming’ procedure that the current semantics document describes). Pat claims that the whole semantics could be re-written that way and none of the current RDF implementations would change. But one can go one step further: there may be different kinds of surfaces (eg, negations) and surfaces can have a name (a bit like named graphs) and all can be put together to provide a powerful semantics for these entities. His further claim was that such an extended semantics of RDF could be powerful enough to describe, conceptually, RDFS or even OWL, ie, the semantics should not be layered any more.

No way I would accept all this argumentation on face value:-), so I have to think about this and, mainly, read whatever Pat may want to write down to understand it. In the meantime, I may have to look into the concepts of conceptual graphs, and the Peircian notation of logic that Pat referred to as inspiration…

A more general take away (see also Chris’ remark above): maybe it is time to look into RDF again? A scary thought. Touching to something that is fundamental on the SW has to be done with extreme care… We will see.

There were two papers in the same session that were very close in subject and topic: one of Jesse Weaver and Jim Hendler on the parallel materialization of RDFS graphs and the one of Jacopo Urbani et al on using MapReduce for RDFS reasoning. (Sigh…, this is where I would like to put a reference!) Both aimed at similar challenges, namely the materialization of RDFS inference results of a graph using parallel computing methods. And there was one more similarity: both had some sort of a classification of the rules in the rule set described in the RDF Semantics document to help improving the processing. (Eg, to analyze which rules should be duplicated among processing nodes and which one can be handled without, or which one need a special treatment for a map-reduce pair). It seems that it would be worthwhile to see if some of these classifications (‘ontology rules’ and the like) could be extended to OWL 2 RL (Jesse Weaver told me afterwards that they want to look into this).  But, to put things into perspective: we are the points when billions of triples can be expanded with relative ease. Who would have thought a few years ago? There was also a remark on one of Jesse’s slide (I do not remember the exact wording) which said that RDFS is insanely parallelizable:-)  It was a really interesting session.

The SW in use session included  a paper from Landong Zuo et al “Supporting multi-view network analysis to understand company value chains”.  Integrating a bunch of data in the UK on companies, integrating them in an RDF store, and let users get information on the ‘value chains’, ie, how companies relate to one another as producers/consumers. Technically, the interesting point was the fact that users had the possibility to interactively add new relationships, new classifications to the system, essentially new rules that could be evaluated. The whole system seemed to be a really cool, a well engineered and well functioning machinery. As the speaker put it, although all conclusions drawn from the system could be found by the users by analyzing databases, but it would take weeks to do what this system can give them in a few minutes. This is exactly the kind of message we need for the outside  world about the usefulness of Semantic Web technologies.

On another session Martin Szomszor presented an experiment they conducted at the ESWC conference, combining RFID-based personal badges with an underlying SW system. The resulting system could be used to show personal contacts among delegates, could help people find others with similar interest, could retrace later whom one met at what point (“I remember talking to that chap, but I do not remember his name!”), etc. Lots of privacy issues, for example, but I would have liked to see that in practice, that is for sure!

Stéphane Corlosquet’s presentation on SW and Drupal was really exciting. I already knew about the plans of Drupal 7 to incorporate RDF management from the start, that all Drupal 7 pages will be annotated via RDFa. The RDFa community has been  fairly excited about that for a while now. But the work done by Stéphane and others provide some additional modules that makes it easy to add a SPARQL endpoint to a Drupal based site easily, to import other RDF content, or to manage the vocabularies used on the pages and the like. They already have such a system running with the current Drupal, but these modules will become part of the standard Drupal 7 module set that one can download from the drupal site. And that is cool.  It significantly lowers the barrier to build Web sites that are prepared to be part of the Linked Data cloud, even if the system administrators are not SW experts. I expect this to open up quite a lot of possibilities…

Off to the next day! More paper and the presentation of the Semantic Web challenge finalists…



  1. i don’t know about ISWC papers in general, but Urbani et al paper you talk about is available at http://eyaloren.org/pubs/iswc2009.pdf.

    Comment by Valentin — October 29, 2009 @ 0:00

  2. Glad to hear that you found our paper interesting (the “Parallel Materialization” paper).

    Just a quick note. It became apparent to me that many of the ISWC attendees were not familiar with the term “embarrassingly parallel.” This is actually a technical term from the parallel computing community (and not, as some may have assumed, a term that I invented). An embarrassingly parallel computation is “the ‘ideal’ computation from a parallel computing standpoint–a computation that can be divided into a number of completely independent parts, each of which can be executed by a separate processor” [1]. So, the main take-away from my presentation is that the RDFS rules can be applied to an abox partitioning (as defined in my paper) in an embarrassingly parallel fashion (i.e., independently on each partition) to produce the RDFS closure.

    Also, as soon as springerlink makes their version of my paper available, I will (be legally allowed–per the copyright agreement–to) publicly post a copy of the camera-ready version at my website, .

    [1] Wilkinson, B., Allen, M.: 3. In: Parallel Programming: Techniques and Applications Using Networked Workstations and Parallel Computers. 2 edn. Prentice Hall (2005)

    Comment by Jesse Weaver — November 1, 2009 @ 23:51

  3. Thanks for asking me to clarify at the end of the presentation. I admit I should have made this more clear. Note that the modules we presented are already available for Drupal 6! I have posted more details on a follow up blog post on the status of RDF in Drupal.

    Comment by Stephane Corlosquet — November 3, 2009 @ 21:53

RSS feed for comments on this post.

Blog at WordPress.com.

%d bloggers like this: