Ivan's Blog

Elie Sanchez was nice enough to send me a copy of the book he edited on fuzzy logic and Semantic Web[1]. Lots of stuff there which are not easy to digest for the non-initiated (and I am not one of those either…). But it is interesting to have a look at it nevertheless. Incorporating fuzziness into the Semantic Web world has been an issue for a while; I have received a number of comments from application developers that such functionality is really necessary to describe their knowledge base. (A relationship between two biological experimental results, for example, can rarely be stated with absolute certainty…).

The book includes several papers that extend the traditional DL with fuzzy statements. In one of the models one can state, for example:

<a:C ≥ 0.6>

meaning that "a" is of class "C" “on a 0.6 degree” . The interesting thing is that U. Straccia gives a full definition of SHOIN(D)’s fuzzy version (SHOIN(D) is the DL variant for OWL). However, as he says, there is no calculus yet for checking the satisfiability of fuzzy SHOIN(D). Of course, this is not the first time these issues are looked at. There was a workshop at last year’s ISWC[2] which addressed similar issues but, somehow, went largely unnoticed; I have not seen any reference to those papers in [1] although, for example, Stoilios et al. reported on a fuzzy version of a (weaker) DL dialect already[3] at that workshop. Oh well…

All papers I saw on the subject in [1] look at fuzziness on an ontology level. I was also looking for something like that on a lower level, ie, have the possibility of adding fuzziness on the RDF level already. Indeed, one should try to keep it simpler and use ontologies only when one really needs it… There is a paper in [2] from Mazzieri and Dragonidoing just that[4]. In their structure one can add a fuzziness on a triple level, ie:

n: s p o.

means that the (s p o) relationship holds on a degree of ≥n. The nice thing in [1] is that all the notions of RDF and RDFS can be translated to this setting, with a reasonable alternative definition of the whole model theoretical semantics of RDFS. They even report to have implemented a forward chaining fuzzy RDFS reasoner in Sesame… Of course the picture would really be complete if this could be expressed in SPARQL, too, but that may not be so complicated. (Hm. That may not be true. It seems that the current model theoretic semantics used in the latest version of SPARQL leads to a number of problems, so a fuzzy semantics might make it much worse. Let us have SPARQL finished first before going there…)

Nice things to think about; it is still to be seen whether this approach will solve real world problems. But it is worth investigating. Looking for a PhD topic? 

(B.t.w.: there will be a similar workshop at ISWC this year. I say similar: probabilistic reasoning and fuzzy logic, though attacking similar problems, are two distinct camps, and the practioners of both fields are very sensitive not to be mixed up with the other…

References

1. “Fuzzy logic and the Semantic Web”, E. Sanchez (ed.), Elsevier, 2006
2. Workshop on Uncertainty Reasoning for the Semantic Web, International Semantic Web Conference, eds, P. Costa, K. Laskey, K. Laskey, M. Pool, 2005. The full proceedings is also available on-line.
3. “A Fuzzy Description Logic f-SHIN’’, Giorgos Stoilos, Giorgos Stamou, Vassilis Tzouvaras, Jeff Pan, Ian Horrocks, in [2].
4. “A Fuzzy Semantics for Semantic Web Languages”, Mauro Mazzieri, Aldo Franco Dragoni, in [2].