Alignments for data interlinking: Call for contributions

Currently the linked data graph is linked at the data level (directly by addressing other resources URIs, through properties or through sameAs). Because the task is often massive, resource publishers tend to write programs for extracting these links, we call them 'link generators'. Yet, at the moment, a few tools exist which each use their own input format, that we will call 'link generator specification', or specification for short.

We would like to investigate the idea of publishing, as a kind of ontology alignments, these specification on the web. Note that the links themselves are already available through source providers or services like sameAs.org. Publishing a high level specification of these links may have many interesting applications:

• Trying to generate direct links by composing alignments (from a to b and b to c),
• Trying to generate more complete alignment by aggregating (several alignments from a to c, maybe obtained by composition),
• Checking the coherence of these link generators at the ontology level (that an entity t, will be mapped to the same entity using different paths of alignments),
• Checking that among published links there are none which would violate such specification (Person is usually sameAs other Person, but some of them are mapped to Organisation);
• Sharing alignments on the web, as well as in resources like Cupboard or the Alignment server.

For that purpose, it is necessary that these alignments or specification could be expressed in a common language which is:

• publishable on the web,
• able to express relations at the ontology level,
• able to express the precise generation of links.

We would like to run a quick experiment during the next summer in order to evaluate primarily the feasibility of this approach and eventually the adequacy of the available formats. Our plan is the following:

• collect as much as possible such link generator specifications or alignments;
• refactor them (possibly automatically) in a common format;
• experiment with this in order to show/evaluate the announced benefits (or others).

So, this is where we need you: we would like every possible resource publisher in the LOD to send us a reference to their link generator specifications. We know that this will not be homogeneous nor easy to understand, but we want to confront ourselves with all this diversity. So we do not ask you to change anything, we take everything: Specialised link specification languages (e.g., Silk-SSL), RDF, SPARQL, XSLT, Perl, Java, C++, etc.

Of course, we would like to make this as open as possible and at least to return the results to the community. We have set up a web site (http://melinda.inrialpes.fr) about this initiative so that we can report on it, but we naturally prefer that all resources be as distributed as possible (alignments are far lighter than the data).

How to participate

• Send an email to: Francois:Scharffe#inria:fr
• Before the end of June
• Link generator specifications
• Any format
• Better if everything is available on the web
• That's it.

Resources

• Silk Link Language Specification: SILK-LSL
• ODD linker: STRINGER project (SQL queries)
• MusicBrainz to Jamendo linker: LD mapper (Prolog code)
• Knofuss (OWL)

References

• Expressive alignment format
• Silk-SSL
• Mediaspaces mapping
• SPARQL++
• ODDlinker
• RKB Explorer
• OKKAM
• Knofuss architecture