Call For Papers
|This workshop is partially supported by EASTWEB|
We recently heard about a telecom project that required reasoning about
10 billion RDF triples in less than 100 ms. This use case was defined
around generating revenue streams through new context-sensitive and
personalized mobile services. This is just one example of a general
demand for reasoning over extremely large scale semantic repositories.
The Web has made tremendous amounts of information available that should
be processed based on formal semantics attached to it. The Semantic Web
community has developed a number of languages (RDF, RDF Schema, OWL)
that deploy logic for this purpose. Impressive progress has been
made on scaleable storage, querying and inference for these languages,
and they are succesfully being deployed on large intranets and
medium-scale web-applications. However, it is unlikely that these approaches
will scale to the amount of information and setting the Web is providing.
The main open question that will be discussed at the workshop is:
Why isn't reasoning scaling for the Web and how can this be fixed?
Some of the limiting assumptions underlying current reasoning languages for
the Web are:
- sets of axioms and facts are restricted in size
- the axioms and facts are static and known in advance
- the inference process must be sound and complete
Each of these assumptions (and most likely others) needs to be revisited
and adopted to the reality as it is provided by the Web.
Contributions form many very different scientific fields will be needed
to establish the paradigmatic shift that is necessary to obtain this
goal. The workshop aims to attract participants from scientific fields
not typically seen at Semantic Web events:
- Economics can tell us about cost-benefit trade-off models, sunk-cost
theory and the role of negotiation in obtaining near-optimal results
under bounded resources
- Cognitive Science can tell us about strategies of human memory and
human reasoning which is so succesful in reaching good enough
conclusion in limited time-spans, using methods such as priming,
attention scoping, recency-based self-optimising memory, etc.
- Computational Learning Theory has contributions to make with their
notions of Probably Approximately Correct (PAC) computing,
and strategies for abstraction and compression of information.
- Combinatorial Search has made recent breakthroughs in handling massive
search-spaces with heuristics based on Monte Carlo simulations.
Some of the ideas that sparked off this workshop can be found in: Unifying Reasoning and Search to Web Scale, Dieter Fensel and Frank van Harmelen, IEEE Internet Computing, March/April 2007 (Vol. 11, No. 2), pp. 94-96