Ontology Mapping and the Semantic Web

Harry Halpin, <H.Halpin@ed.ac.uk>
<harry@w3.org>

Textorized Parthenon

Ontology Mapping and the Semantic Web

Links Join Across Different Sources of Information

 Links between column headings

RDF (The Semantic Web) is: subject/predicate/objecting using URIs for subject and predicates, and maybe objects.

The Problem

The Semantic Web uses URIs not just to refer to web-pages, but to things.

How can we determine what a URI identifies? Can it be determined from the information found in the associated description of the URI or the use of the URI in other RDF graphs? How can we build semantic contientients built on shared URIs rather than isolated semantic islands?

Two Strategies

  1. Try to get people to re-use URIs: People have to be able to find URIs and then applications have to help people re-use them?
  2. Try to get URIs for the "same" concept to match: State-of-the-art here still is basically broken.

The Semantic Web is not AI

  1. The Semantic Web does not require a large, general-purpose, common-sense ontology.
  2. The Semantic Web does not need to understand most humanly relevant contents. as it lacks the ability to perceive and interact with a physical environment.

The Semantic Web has humans to interact with the world for it, and can parasite off their embodied knowledge

It's collective intelligence (CI), complementing human intelligence by creating an external knowledge representation for Planet Earth, not articial intelligence, i.e. a simulation test of the utility of internal knowledge representation in building something that will pass the Turing Test.

Beyond Toy-Systems

Most work on ontology matching, including work on the Semantic Web, has taken place in restricted domains with little actual data, ala Winograd's block-world.

While ontology matching systems seem to work well in these limited contexts, the general rule of thumb is that outside their specific development environment their performance is always poor: 30 percent.

Thus giving ontology matching in general a very poor reputation amongst actual developers of Semantic Web applications.

Currently using manual ontology matching via owl:sameAs, but already this is causing problems: http://dbpedia.org/resource/Berlin owl:sameAs http://sws.geonames.org/2950157/

Linked Data Ecosystem

With the advent of Linked Data, the Semantic Web has now moved beyond toy systems to over 10 billion bits of re-usable information in RDF, ranging from the BBC to Wikipedia in RDF. No excuse for toy systems.

Linked Data

What needs to be matched?

Question:Is there too many or no URIs for things?

Answer: : Sample the Semantic Web using a query log

Data Set: Microsoft Live Search Query Logs

Too Many URIs

There are probably too many URIs for the same thing...

An average of 1,339 URIs (S.D. 8,000) returned per query

That's a lot, but most may obviously be URIs that just mention the term...

Saul Kripke

Wrapping Up

RDF is now part of Yahoo's Open Search Strategy and is rapidly being adopted by Microsoft. Google has admitted not doing well with structured data and is now after "semantic search"...so these problems are real and being encountered by industry. Yahoo!

Any questions?