This means that even in the *current* Linked Data web, the annotation aspect is not “linked” at all: you cannot discover the annotations of dbpedia:France by links alone. The nature of the URIs used does not affect this; it merely determines if you need an additional owl:sameAs step to interpret the annotations.

Good point re: simply adding data at someone’s URI space. Linked Data publishers could usefully provide or enable this kind of feature, suitably qualified for trust-worthiness, as part of their publishing framework.

btw, I love the “web as a cms” concept, I think its a really important change in perspective for how organisations can engage with the web.

Thanks for the comments, you make some good points. I’d argue that there’s still not enough of an understanding that creating new URI aliases and using owl:sameAs is not the only way to publish RDF data. I see little advice to the contrary.

Your point about surfacing data from existing applications is well taken, creating new URIs based on database keys may be the simplest approach. Although at some point you’d expect there to be some alignment between URIs in, e.g. dbpedia. So starting from a perspective of “tagging” a local identifier with a public URI is equally valid.

The other thing to highlight is that some organization could have a valid and important role in the development of the web of data if they only focused on defining stable URIs for things — governments might do this for example, enabling citizens and businesses to use these as points for common annotation. To some extent this is what is happening in the first phases of the development of the UK government linked data.

Thanks Leigh, great essay. It would be unfortunate if publishing data in RDF becomes an exercise in bondage-and-discipline, which is what some linked data advocates (the new neats?) seem to advocate.

Horrible antipattern I’ve seen rather too often:

owl:sameAs dbpedia:Sesame_Street ;
rdfs:label “Sesame Street”@en ; ….

First: The only reason why we can have this conversation today is because now we have quite a nice selection of existing, reasonably stable, well-managed, dereferenceable identifiers around, from DBpedia, Freebase, and hopefully soon from many governments and other public organisations. This wasn’t the case two years ago. Two years ago, the conversation was: “Use a bNode, or use some unresolvable URI, or use the identifier for France that someone minted in their FOAF file, or mint my own.” Out of those, minting your own is clearly the best.

Second: Until quite recently, there was a widespread attitude that URI aliases is harmful, that the Semantic Web would be impossible to attain if we invent new identifiers rather than going to great lengths finding some existing canonical identifier. This attitude is slowly disappearing, luckily.

Third: If the backend system that you expose as RDF has its own mechanisms for identifier management, then you’re really better off publishing your data attached to URIs that are based on these internal identifiers. Linking to other identifiers, which are managed independently, should be a separate step; especially if you use imperfect heuristics to discover the coreference.

Fourth: If your data is inherently connected to someone else’s identifiers (e.g. because you annotated their identifiers in the first place, or because both of you use the same non-URI identifiers as a base, like motorway numbers), then I think it’s reasonable to simply publish RDF documents that use those foreign identifiers.

Fifth: If you trust the owner of those foreign identifiers to be around longer than the expected lifetime of your dataset, then it’s quite fine to base your data on their identifiers. If not, then you should manage your own identifiers and link to theirs.

I agree with you, but the usual discussion of how to publish Linked Data — and I’m as guilty of this as anyone — doesn’t explore notions that the URIs that we use to name things (Principle #1) may be in other domains. Or that “looking up” a URI (Principle #4) doesn’t have to be a direct dereference or that URI, it may involve looking up that URI using a service like SameAs.org.

Its natural enough when putting together tutorials and frameworks for linked data publishing to focus on the mint your own URIs & publish from those URIs approach, because that gets the core message across. Its also natural when there aren’t many stable URIs to reuse.

But as things grow, the emphasis will natural change, and so as we encourage people to adopt the Linked Data style of publishing data, we need to encourage more annotation. Its not only simpler, its also side-steps a few issues, as I alluded to in the post.

I’m also being a bit playful with the whole “heresy” thing.

Cheers,

L.